• High-flying IPO, high-flying hype: Reuters reported that the SpaceX IPO is 3.3x oversubscribed as of yesterday, but I wouldn’t believe a word from anyone buying or selling into this IPO. When Google announces it will pay $920 million per month to rent xAI’s idle GPU capacity just 5 trading days before the IPO, you can smell the desperation. Google doesn’t need GPUs, but it does need exit liquidity, as it owns a 5% stake in SpaceX with a 17,900% RoI.

• The Street wants you believe that AI will be the growth driver over the next 5 years, but this implies that Grok can overtake frontier market leaders like Google, OpenAI, and Anthropic. Musk is hellbent on doing this, so there’s no way to avoid huge cash burn until he can achieve this.

• Beware of the options market on a purposefully executed “low-float, high-valuation” IPO, meant to create both a passive flow squeeze, and a huge gamma squeeze after SpaceX’s options market opens on day 4 or 5 after the IPO.

• Beware of early, unlocked selling if you’re long SpaceX, and hope that the Vested Interests “sway” the share price higher—the way they do with Tesla—after passive buying ends on day 15 post-IPO. Musk’s “tiered” unlocked selling schedule starts the day after Q2 results, and will allow 13%+ of early SpaceX investor shares to be unlocked to sell.

• Realize that both Anthropic and Google can begin exit procedures from their GPU leasing agreements with xAI as of December 2026, which means that earnings could return to losses from Q1 or Q2 2027.

• The “Telsa-SpaceX” merger speculation has almost become consensus. Assuming this is Musk’s game plan, it will take at least 3 to 4 quarters of regulatory reviews, and you can bet that the Tesla bagholders will start an activist revolt to demand much higher merger ratios, which would prolong the event even further. See my deep dive into SpaceX’s IPO here, and why it makes more sense for SpaceX to cancel its IPO for a 1:1 merger with Tesla here. For my take on SpaceX’s S-1 filing, see this report here.

====================================================================

When Jamie Dimon, JPMorgan’s iconic CEO, asked Elon Musk at an investor meeting last week why SpaceX is going public now, Musk smiled and replied, “We’re embarking on a massive new growth phase, and we need capital for that.”

This isn’t exactly true. In fact, it’s as close to a lie as one can get without lying, given that the $75 billion of funds that will be raised by SpaceX’s IPO will go to pay down liabilities linked to Musk’s failed AI start-up, xAI:

• xAI, was burning cash at a rate of over $1 billion per month last year, with minimal revenue. While segment cash flows aren’t disclosed, the EBITDA less capex in SpaceX’s AI division last year came to a $15.5 billion deficit.

• On top of this cash incineration, xAI has huge investment needs: including supplemental capex, xAI saw total investment outlays of $23.4 billion in Q1 2026, with only $0.8 billion of revenues, eclipsing last year’s total of $19.8 billion of capex, on $3.2 billion of revenues. Most of this was GPUs and equipment needed to build out 2 massive data centers for xAI.

• xAI and Twitter were drowning in debt that carried 11.4% interest rates and required nearly $2 billion in servicing costs last year. SpaceX replaced this in March with a $20 billion bridge loan, at a much more reasonable rate of 4.6%. Nevertheless, including $10.2 billion of high-yielding AI-related financing that xAI used to carry it through Q2, 40% of SpaceX’s $75 billion IPO will go to pay down its AI debt, while much of the rest will be needed to cushion the cash burn for the rest of the year. Goldman Sachs estimates that SpaceX will burn through $350 billion of cash through 2030, or $70 billion per year on average, before generating positive cash flow.

If Goldman's cash burn estimates sound aggressive, think about it this way: on user enjoyment benchmarks, xAI ranks 4th among frontier labs, but on tests that enterprise buyers actually care about, rigorous reasoning and autonomous task completion, it ranks last (see Figure 1). That gap is what the immense cash burn is funding: a bet that SpaceX can leapfrog Google, OpenAI and Anthropic to reach 1st place. From this standpoint, Goldman’s estimates of $350 billion in AI cash burn through 2030 may be slightly optimistic.

Figure 1: Frontier AI Lab Rankings by Benchmark (June 2026)

Sources: Stanford HAI AI Index Report 2026; LMSYS Chatbot Arena; LLM Stats; Scale AI SEAL Leaderboard; ARC-AGI Prize leaderboard. Rankings are publicly available scores as of June 2026 and can change as new models are released.

In light of this, SpaceX could run through its $75 billion IPO proceeds as soon as Q1 or Q2 next year, especially if both Google and Anthropic cancel their GPU rental deals with xAI, which appears to be the case (more details below). And this is before factoring in Musk’s ambitions of building a “Terafab” to build cheaper chips than TSMC, which would cost upwards of $60 billion just for the initial phase. This is why it’s hard to believe that Musk would’ve taken a debt-free, highly profitable SpaceX public if its “Frankenstein” of an AI division didn’t exist.

IPO is only 3.3x Oversubscribed a Day Before Books Close

Reuters reported yesterday that bookbuilding so far has reached $250 billion, which is only 3.3x oversubscribed, with only a day left before books close after today’s market close (link here).

While various press reports, including the Reuters article, say this is an indication of “strong” demand, given the record funds being raised, it seems a bit underwhelming in light of all the hype since January and 2 extra days of marketing (the roadshow launched on June 4th rather than June 8th, as the SEC finished their review earlier than expected).

Also, looking at the top global IPOs in Figure 2 (which shows inflation-adjusted funds raised), it’s clear that, while the deal size of IPOs like ICBC and Meta may have been smaller, their huge oversubscribed rates of 45x and 25x, respectively, show how much more popular they were than SpaceX.

Figure 2: Top 10 Global IPOs Ranked by Oversubscribed Rates

Source: Bloomberg and various press reports. * Note: All values are inflation-adjusted for today’s prices. ** Also note that Price/Sales multiples use trailing 12-month revenues.

Moreover, Musk clearly had Google step in to sign a last-minute deal with xAI, in a shameless attempt to pump the IPO, which was only 2x oversubscribed last week, and not a good sign even after its $75 billion size is considered.

Google GPU rental deal: After last Friday’s market close, SpaceX announced a deal with Google, where xAI rents 110,000 of its GPUs at its Colossus 1 data center to Google for $920 million per month, or $11 billion per year through 2029.

This is exactly half the size of a similar deal announced with Anthropic on May 6th (220,000 GPUs for a monthly fee of $1.25 billion, or $15 billion per year), and combined, would lead to an annual revenue stream of $26 billion, which is nearly 40% higher than SpaceX’s total 2025 revenues of $18.7 billion.

With an estimated operating margin of around 65%, this could have turned last year’s consolidated operating loss of $2.6 billion at SpaceX into a profit of $14.3 billion, which is quite significant.

But this kind of announcement, just 5 trading days before the largest IPO in history, is actually as unserious as one would expect, once you look at the details:

• Google doesn’t need xAI’s GPUs: Google described this deal as "bridge capacity to meet surging customer demand for our agent platform, Gemini Enterprise." Bridge capacity, by definition, is temporary, and the deal term says that, after December, Google or xAI can cancel the agreement with 90 days’ notice. But why announce this a week before the IPO?

• Google owns a 5% stake in SpaceX: After investing around $500 million in SpaceX in 2015, when SpaceX was valued at only $10 billion, Google’s current stake is worth nearly $90 billion at the IPO price of $135/share. This is a 17,900% RoI for Google. It should also be of no surprise that most queries about SpaceX on Gemini 3 lead to highly positive takes on the IPO, unlike replies to questions about Tesla robotaxis versus Waymo, which are much more objective.

The Anthropic deal was also walked back by Musk, who commented that the $1.25 billion monthly GPU rental agreement with xAI “is a 180-day lease with 90-day notice mutual cancellation. The short term was our request, not Anthropic’s.” This led to the S-1 filing being amended to match Musk’s statement, so Anthropic’s deal is on a “rolling 90-day schedule” from August, as it has its own AI infrastructure coming online by the following schedule:

• Q4 2026: Anthropic’s AWS capacity arrives—first alternative to xAI’s GPUs.

• December 2026: Over 1GW of Google TPU capacity comes online.

• December 31, 2026: Google’s 90-day termination window opens on the xAI contract.

• Q1–Q2 2027: Google/Broadcom TPU capacity begins flowing to Anthropic.

The 1GW+ of Google TPU capacity arriving by the end of 2026 is sufficient on its own to replace the inference workloads now running on xAI's 220,000 GPUs (TPUs are said to be perfect for this type of large-scale transformer inference). Combined with the existing AWS Trainium capacity already running, Anthropic won’t need xAI’s GPUs before the end of December.

Analysts Struggle to Make Estimates Fit Valuation

Because of Musk’s insistence on the nose-bleed valuation of $1.77 trillion for the SpaceX IPO, analysts at the investment banks that are underwriting the offering are struggling to make the maths in their estimates work.

And while Starlink in the Connectivity Segment is the crown jewel of SpaceX’s growth and profits (EBITDA grew by 86.2% YoY last year and made up 109% of total 2025 EBITDA), the analysts at Goldman Sachs and Morgan Stanley are pouring their efforts into showing how much the cash-burning AI Segment will grow.

Goldman Sachs is the lead underwriter, while Morgan Stanley is playing second fiddle, which is surprising given how much more of a Tesla cheerleader Morgan Stanley has been. Both made headlines last week as a few of their estimates were leaked to the press, and they’re notable, just to see how far these banks are willing to go in order to flog more SpaceX shares to their clients:

2030 vs 2025 Revenues

• GS: $18.7 billion—>$474 billion (+25.4x)

• MS: $18.7 billion—>$330 billion (+17.7x)

2030 vs 2025 AI Segment Revenues

• GS: $3.2 billion—>$322 billion (+100.6x)

• MS: $3.2 billion—>$190 billion (+59.4x)

2030 vs 2025 EBITDA & EBITDA Margin

• GS: $6.6 billion—>$352 billion (+53.5x); EBITDA margin: 35.3%—>74.2%

• MS: $3.2 billion—>$230 billion (+34.9x); EBITDA margin: 35.3%—>69.7%

Assuming SpaceX net debt is flat by 2030, despite the AI segment’s cash furnace, and the listing price of $135/share, Goldman’s estimates lead to the trailing 12-month EV/EBITDA of 267.6x coming down to 5.0x by 2030, while Morgan Stanley’s assumptions get you to a 7.7x multiple.

Again, to assume that xAI can disrupt the incumbent frontier market leaders like Google, OpenAI, and Anthropic over the next 5 years without taking bigger hits to earnings is a bit hard to imagine: xAI has spent over $25 billion on GPUs since 2023, by my estimates, and the first batch of these will need to be replaced by higher-performing new GPUs on the market from the 2H of 2027. Figure 3 below compares SpaceX’s earnings results to what they would be without its AI Segment, just to drive this point home.

Figure 3: SpaceX Earnings With & Without its AI Segment

Source: SpaceX & MH estimates. Ex-AI Segment numbers assume flat net interest expenses, and a 21% effective tax rate.

Musk’s Engineered “Passive Squeeze” and the Wall of Insider Selling

There’s been a lot of controversy over Musk’s push to get various indexes to scrap their usual rules on inclusion for newly listed companies, and grant SpaceX early inclusion. Among the larger exchanges, the “bend the knee to Elon” crowd includes MSCI, the Nasdaq 100, Morningstar/CRSP, and the FTSE Russel 1000.

From day 5 after the IPO through day 15, these are the key indices that will lead to funds that track them to buy SpaceX shares, whether they like the stock or not, in one trading session (except for CRSP, which will spread its estimated $1.8 billion of buying over 5 days), which will create a “passive squeeze”, and likely push the stock higher.

True to his moneygrubbing style, Musk has purposefully limited the amount of SpaceX shares to be offered in this Friday’s IPO to only 4.3% of total shares outstanding, in order to execute a “low-float, high valuation” strategy that will push SpaceX shares higher. By comparison, megacap tech stocks like Meta (Facebook) listed 15.4% of their shares when they went public in 2012, while Alibaba floated 13% of its shares in its 2014 IPO.

S&P 500 Doesn’t Bend the Knee to Musk

But Musk finally met with some karma, however, when the S&P Index Committee announced on June 4th that they wouldn’t change any of their inclusion rules for SpaceX. This was a pretty big slap in the face, as passive flows are estimated to have been around $40 to $50 billion, versus SpaceX’s float of around $75 billion, which would likely have led to circuit breakers going off throughout the trading session.

Given the 12-month seasoning period in SpaceX trading which is required by the S&P Index Committee, it likely won't have a chance to be considered until June 2027 at the earliest, and must also show 4 quarters of cumulative GAAP net profits, with the final quarter being in the black. Unless Anthropic and Google are still renting GPUs from SpaceX (highly unlikely, given both won’t need xAI’s GPUs after December), it’s hard to see SpaceX being profitable next year.

Figure 4 below shows the main indexes that will have an impact from day 5 after the June 12th IPO, and day 15, when the Nasdaq includes SpaceX. In total, there should be around $14 to $15 billion of passive flows during these 7 trading sessions, which doesn’t sound like much when you consider that Tesla trades on average around $22.5 billion per day.

However, Tesla’s free float is around 83% of its shares outstanding, so its daily traded value amounts to only 1.8% of its float. For SpaceX, the $15 billion or so in passive flows between day 5 and day 15 amounts to 20% of its float. For the far right column in Figure 4, I assumed that the daily turnover will initially be around $27 billion, which is 36% of its float, but it could be higher given the high volatility (details below).

Figure 4: Key Indexes for the Day 5 to Day 15 “Passive Squeeze”

Sources: Nasdaq Index Metrics; MSCI Global Standard Index Early Inclusion Notice (June 8, 2026); CRSP Market Indexes & FTSE Russell IPO Inclusion Frameworks; Bloomberg Intelligence & BNP Paribas Liquidity Flow Analysis. *Note: CRSP is spreading its inclusion of SpaceX over 5 trading sessions. Also note that trading volume is estimated to be $27 billion per day.

Gamma Squeeze From SpaceX Options Market From Day 4 or 5

Anyone who’s traded Tesla before is keenly aware of how much the stock is “swayed” by the Vested Interests via the options market, namely huge volumes in weekly calls with near-the-money strikes. It seems like SpaceX will be like Club Med for the Vested Interests, given its small float, which Musk’s financial advisers surely flagged to him. Expect the following trends:

• High implied volatility of over 100% during the first week or two, due to market makers finding inventory, etc. with such a small float, and wide bid-ask spreads. This will likely drop into the first quarterly earnings results (Q2 2026) some time in late July or early August.

• Gamma sensitivity during the first several weeks could lead to mayhem in light of the expected high allocation to retail investors (who might leverage more by use of call options), and SpaceX’s thin float, which will exacerbate moves in the stock.

Don’t Forget About Exit Liquidity: Early Tiered Selling

Figure 5 is a nice visual that help explain another unique aspect of SpaceX’s IPO, which is its strange schedule for early investors to be unlocked and allowed to sell.

Usually, there’s a 180-day lockup period, but because there are early investors who’ve been waiting for over 12 years, Musk provided the “tiered” unlock schedule in Figure 5, which has a “Pool A” for early investors (private equity, venture capital, etc.), and a “Pool B” for insiders and executives, which ends 366 days after the IPO. Here are the key points:

• “Friends & Family” tranche is 5% of IPO offering from day 1, which could weigh on the share prices in the absence of passive flows from day 5 after the IPO, and the options market not established yet. Assuming this is worth $3.75 billion of selling—if they all sell—it could amount to 10-15% of average daily volume until passive flows and call buying begin.

• Selling of 20% of one’s holdings will start the day after Q2 2026 results, which should be in either late July, or early August.

• If SpaceX trades at or 30% above its IPO price of $135 in 5 of any of the 10 days before Q2 earnings results, early investors can sell an extra 10%.

• Roughly 7% of Pool A’s tranche to be unlocked between August 21st and October 25th (days 70, 90, 105, 120, and 135).

• Another 28% of Pool A shares unlock after Q3 earnings, which may sound like a great selling/shorting opportunity, but it coincides with the Nasdaq’s rebalancing in December, which could lead to another $20 billion of passive flows, if SpaceX has a 30% float by November 30th.

• Insider selling begins a year from the IPO, but amounts to 52% of shares outstanding to be unlocked. I wonder what kind of plans Musk’s financial engineers have come up with from that point, if SpaceX is still burning cash.

Figure 5: Tiered Unlocking Schedule for Early SpaceX Investors

Source: SpaceX, Reuters, and MH research.

While the meme below is how active market watchers may view the SpaceX IPO, sometimest the best trade is no trade.

Source: Twitter.

NOTE: NOTHING IN THIS REPORT IS INVESTMENT ADVICE.

[This was supposed to be a quick note, and it is short, but Figures 2 to 4 took an exorbitant amount of research to produce. They show just how dangerous Tesla’s robotaxi fleet in Texas is compared to both its new rivals in Texas and even human drivers. If you’re not in the mood to read much of this on a Friday, do at least take a look at the tables in this report.]

The Texas Department of Motor Vehicles has officially rubber-stamped Elon Musk’s "FSD self-certification" model, granting permits for just 42 Tesla robotaxis under the state’s new, stricter automated vehicle regulations. While the Cult of Musk was rejoicing, the meager fleet size—only 7.2% of Waymo’s Texas presence—suggests this is more about maintaining its veil of self-driving prowess than actually launching a real driverless ride-hailing service.

If Musk were truly confident of Tesla’s FSD being over 7x safer than human drivers, why did he only apply for 42 permits? Waymo obtained 577 permits, Zoox got 35, and May Mobility got 18 AV permits to operate in Texas.

In my May 13th magnum opus on why Musk is under pressure to tone down his robotaxi ambitions (“Why Musk is Walking Back Tesla’s Robotaxi Roll Out”, here), I wrote that Musk would need to self-certify FSD as capable of Level 4 autonomous driving or stop all “unsupervised” robotaxi rides in Texas altogether.

In light of the Texas DMV granting Tesla permits to operate 42 Tesla robotaxis in the state yesterday, it appears that Musk produced documents for the Texas DMV that say Tesla’s FSD is indeed a “Level 4 automated driving system”.

There are reams of data in my May 13th report (here) showing just how dangerous Tesla’s FSD is, and Figures 2 and 3 below provide further evidence, compared to Waymo, Zoox, and May Mobility’s track records, on how far Tesla is from achieving Level 4 autonomy.

While the Tesla fanboys on Twitter were ecstatic over this news (and subjected me to significant vitriol for having been skeptical), it didn’t impact Tesla’s share price: The Texas permit news came out around an hour before Tesla closed yesterday’s session up by only 0.40%, and it declined by around 0.30% in after-hours trading.

It seems that the big Reuters scoop about Tesla’s AI trainers not trusting FSD weighed more on the stock (see Video 1 for a short summary of Reuters’ excellent scoop, with the full article here).

Video 1: Reuters’ Scoop on Tesla’s AI Trainers Not Trusting FSD

Source: Reuters.

Great Part of Reuters’ Scoop on Tesla AI Trainers Not Trusting FSD:

Inside Tesla, as these events approached, staffers worked long hours mapping routes and training the software on specific hazards to make the company’s self-driving technology appear more capable than it really is, four of the former Tesla employees told Reuters. The staffers said these labor-intensive safeguards are impossible to deploy on a broad scale.

The significant points about Tesla obtaining permits to operate 42 of its “robotaxis” in Texas are as follows:

Tesla barely runs its “unsupervised robotaxis”: Figure 1 below shows just how laughable Tesla’s confidence level is about the safety of its “robotaxis”. Musk promised robotaxis for “half the US population by the end of the year” in July 2025, yet Tesla is doing roughly 2 driverless rides per day in just 3 cities in Texas. On average, Waymo does 735x more rides per day, while Uber does 33,333x more daily rides than Tesla’s “robotaxis” in the 3 main cities of Texas.

Figure 1: Proof of Why Tesla Isn’t Serious About “Robotaxis” Yet

Source: Tesla estimates of “unsupervised” robotaxi rides are based on data from the Robotaxi Tracker website; Waymo estimates are based on February 2026 expansion disclosures and NHTSA incident reporting frequency; Uber figures estimated via municipal TNC trip data and market share extrapolations.

Read more

• On February 2nd, the Paris Public Prosecutor’s office (PPO) raised its Cybercrime unit’s probe into xAI and its child pornography crimes to a criminal investigation.

• In the “Risk Factors” section of SpaceX’s S-1 filing (prospectus), they acknowledge that Grok is the subject of regulatory investigations across 8 agencies connected to nonconsensual synthe…

Read more

• The SpaceX IPO has launched despite international liabilities at xAI: SpaceX made its S-1 filing public yesterday, and, despite my note last week about the IPO possibly being canceled and Tesla mooning in response (here), it looks like they’ve overcome any litigation holds due to France’s criminal probe into xAI’s CSAM and anti-Semitic posts. If the IPO roadshow doesn’t start by June 8th, it could indicate that xAI legal liabilities may have stopped it.

• The Anthropic deal is the biggest positive takeaway: The S-1 disclosure that Anthropic has agreed to pay $1.25 billion per month ($15 billion annually) to rent xAI’s idle GPU capacity at an estimated 70% operating margin is a huge boon for SpaceX’s short-term earnings. But this isn’t sustainable, as other hyperscalers like Amazon and Google will surely offer better deals once they have sufficient capacity over the next 18-24 months. The S-1 itself notes Anthropic's Amazon deal includes "nearly 1 GWh of new capacity by the end of 2026," meaning it wasn't fully available yet. Colossus was available now.

• Anthropic deal assures an extra pump in December from passive funds: Given Musk’s strong push for SpaceX’s early inclusion in the Nasdaq index (which has yet to be confirmed), this setup is perfect: it pumps SpaceX’s profits through Q4 2026, which coincides with the Nasdaq rebalance that will lead to another huge passive-buying squeeze as SpaceX’s float rises due to insiders cashing out (see details in this excellent report). But in light of all the capacity coming online—including Anthropic’s—by mid-2027, it’s doubtful this highly profitable rental deal will last.

• Why SpaceX IPO investors will be bag holders of Musk’s failed AI: While the SpaceX S-1 doesn’t break down cash flows by segment, its AI segment’s Q1 EBITDA minus Q1 capex this year is -$8.3 billion, or 3.1x Q1 2025’s -$2.68 billion. The EBITDA minus capex equation is not too far off SpaceX’s stated group free cash flows (refer to Figure 1).

• Tesla is mentioned as a co-developer of “Terafab” and “Macrohard”: The estimated cost of a “Terafab”, or a semiconductor fab with 1 terawatt of annual compute capacity, would cost around $80 billion in the US. On top of that, SpaceX states that Tesla will also be a co-developer of “Macrohard”, an agentic AI platform, which helps explain why Tesla guided for 2026 negative free cash flow (FCF) due to planned capex “in excess of $25 billion”.

• The SpaceX IPO is clearly aimed at retail, which could backfire: There are around 18 pictures of rockets before one can get to the table of contents in the SpaceX S-1 (Facebook and Slack only had around 4 to 8), and the word “Mars” appears 63 times in the S-1. Word on the Street is that the planned valuation is now $2 trillion, up from $1.75 trillion a few weeks ago, and that all underwriters are using their retail arms as much as possible (e.g., Morgan Stanley’s E-Trade). It’s no surprise that Musk decided to allocate 30% of the IPO to retail investors, versus the usual 5% to 10%. The 15% retail allocation for Pershing Square’s April 2026 IPO to retail investors who quickly flipped their positions, rather than “sticky” long-term holders like pension funds, was blamed for the 18% drop below its IPO price by which founder Bill Ackman.

• “Crown Jewel” Starlink isn’t that profitable & faces stiff competition: SpaceX doesn’t include launch costs from its Space segment when disclosing earnings in its Connectivity segment (“Starlink”), which they say had operating margins of 21.3% in 2025, down from 26.4% in 2024. Adjusting for launch costs at an assumption of $17.5 million per launch, however, the 21.3 operating margin reported for Starlink drops to 11.7% when 122 of its satellite launch costs are included. Rivals who are catching up and/or mentioned in SpaceX’s S-1 include Amazon, Eutelsat, OneWeb, Telesat, AT&T, as well as state-backed competitors like China’s Guowang and Europe’s Iris². Starlink’s growth may have peaked out in developed countries, as its aggressive expansion in developing countries—where monthly fees are much lower—led to ARPU dropping by 18% YoY in Q1 2026.

• Nasdaq early inclusion is still not set in stone: The only reason I’d buy into the SpaceX IPO is if it were assured that Musk got the Nasdaq index to grant SpaceX early inclusion, which is rumored to happen on the 15th day after the IPO. Because it would force passive funds like QQQ and many others to buy the day after early inclusion was announced, it would cause an epic “buying squeeze”. Details on how I’d trade this below.

• SpaceX's IPO is aiming for $80 billion, which could tank Tesla’s stock: Because Tesla has roughly 40% of its shareholder base made up by retail investors, if the SpaceX IPO is hyped up enough by Musk and retail Tesla holders see it as more attractive, it could cause a sell-off in Tesla shares (see details in Figure 2 below). If the IPO gets canceled over xAI criminal investigations, Tesla should rise as the prospects of a merger with SpaceX as a private entity increase, as mentioned in my last report.

• Key financial tables are at the end of this report in Figures 3-5, but the link to the complete SpaceX S-1 is here.

===================================================================

Key Points From the SpaceX S-1 Filing

• Anthropic is both a saviour for xAI and its potential slayer: The biggest takeaway from the S-1 is positive: Anthropic agreed to pay $1.25 billion per month to rent xAI’s estimated 800,000 GPUs. The $15 billion of annual revenues from renting xAI’s GPUs to Anthropic should generate an estimated operating margin of around 70%, or $10.5 billion. This is as risky as all of Tesla’s Auto division earnings being leveraged on only the Model Y right now: if Anthropic uses its 90-day cancellation rights on its GPU rentals from SpaceX, $10.5 billion of annual operating profit would evaporate from SpaceX. To put this into perspective, had this Anthropic deal been included in its 2025 earnings, SpaceX’s revenues of $18.7 billion would’ve been 80% higher at $33.7 billion, and its operating loss of $2.7 billion would’ve been an estimated $7.9 billion with a respectable margin of 23.5%.

Figure 1: SpaceX Segment Information

Source: SpaceX & MH research. * Note: EBTIDA = Operating Profit (Loss) + Depreciation & Amortization ex-Impairments.

• Why would Anthropic pay a 70% margin for xAI’s GPU cluster? Speed and scale. Anthropic needed a contiguous, 300 MW/100,000+ GPU cluster immediately to train their next frontier model. Building out a data center of that size takes traditional cloud providers 2 to 3 years due to permitting, power substations, and construction. Anthropic is paying a massive premium to buy time, but this could change as the major hyperscalers often have 18 to 36-month lead times for new, massive power substations.

• SpaceX books GPU depreciation to R&D to boost gross margins: This could be considered an aggressive accounting trick, but it’s also something Tesla has been doing under US GAAP. By booking $2.38 billion of GPU depreciation into Q1 2026 R&D rather than Cost of Goods Sold (COGS), the AI segment looks like it has a gross margin of 44.3%, rather than a negative gross margin of 66.7%. They justify this by claiming the GPUs are being used to research and train Grok, but since the S-1 says that SpaceX has just started to rent its GPU infrastructure to Anthropic, auditors could likely force SpaceX to move this depreciation into COGS. This is a clear sign that Musk is still obsessed with showing higher “gross margins” than operating margins, just like he does at Tesla.

• Shareholders’ equity rose by 13.4x in the 3 months through Q1 2026: Shareholders’ equity at SpaceX was only $2.57 billion at the end of December, but rose to $34.5 billion in Q1 2026. This came on the back of debt and equity financing of $16.1 billion in 2025, and another $22.7 billion in Q1 2026. Some of this debt is quite

• Twitter’s MAU is eroding, and xAI’s is weak: Numbers in the S-1 imply Twitter's monthly active users (MAU) have remained largely flat or seen a slight erosion versus competitors like Threads and TikTok. xAI's standalone consumer MAU for Grok is vastly smaller than OpenAI's ChatGPT, which is exactly why xAI shifted its business model to B2B enterprise infrastructure and cloud rentals.

• Q1 2026 cash burn at xAI is rising massively: Assuming most of the net non-operating losses at SpaceX are attributed to xAI’s debt burdens, its negative free cash flow was over $10 billion in Q1 2026, or 2.9x that of last year’s Q1, and 58% full-year cash burn of $17.5 billion in 2025. The AI' segment’s negative EBITDA of $0.61 billion, less its capex alone, amounts to negative cash flows of $8.332 billion in Q1 2026 versus negative $2.679 billion in Q1 2025.

Key Exaggerations in the SpaceX S-1 Filing

There are outlandish pumps within the SpaceX S-1, including a human civilization on Mars, but the following items are also noteworthy exaggerations of the outlook at SpaceX that have a more immediate impact:

• Overstating Starlink’s profits by 83%: On a pro-forma basis, SpaceX states Starlink’s 2025 operating profit at $1.19 billion, which is a 15% YoY rise from 2024’s $1.03 billion. However, because Starlink excludes launch costs of around $17.5 million, its 2025 operating margin of 36.5%, instead of an actual 11% margin if launch costs were included.

• Pitching xAI as a “high-margin” business: SpaceX heavily conflates “AI application revenue” with what is actually just wholesale data center leasing. They pitch the AI segment as a high-margin software/AI agent business, but the reality is that its main source of income will be from Anthropic’s rental revenue from using xAI’s Colossus 1 and Colossus 2 AI data centers. Furthermore, their Connectivity “TAM” (Total Addressable Market) relies on capturing 100% of the theoretical global Direct-to-Cell market, which ignores terrestrial telecom monopolies and regulatory spectrum blockades.

• Overstating xAI’s TAM: The S-1 touts a multi-trillion-dollar AI TAM, but its immediate success relies entirely on scaling physical hardware (e.g., GPUs, power units, cooling expenses, etc.). The flaw is the assumption that power grids and liquid-cooling supply chains built by xAI will scale their actual capabilities of achieving artificial general intelligence, or AGI. SpaceX also brushes over the risk that, if AI model efficiency improves drastically (e.g., SLMs replacing LLMs), the demand for massive centralized compute clusters could crash.

• Starship to begin launching V3 satellites in 2H 2026: The S-1 aggressively targets the 2H of 2026 for a regular cadence of operational Starship launches carrying the massive next-generation Starlink V3 satellites. This is the hardest proclamation to believe in the outlook section for each of SpaceX’s segments in the S-1, in light of Starship having crashed for its 11th time last week. Going from test flights that crash each time to successful launches with big payloads in the next 6 months is unlikely. It also assumes FAA regulatory clearance for a massive amount of launch volumes, which may not be possible.

Starlink is Risky at Over 60% of Profits

On a pro-forma basis, Starlink is only 38% of SpaceX’s 2025 gross profits, but adjusted for the launch costs that SpaceX excludes from that segment (roughly $17.5 million per launch, or $2.135 billion in 2025), its weight of gross profit drops to 14%.

Starlink’s monthly ARPU (average revenue per user) dropped from $86 per month in Q1 2025 to $66 per month in Q1 2026 and from $91 per month in 2024 to $81 per month in 2025. This was due to Starlink having expanded more in lower-priced emerging markets than in mature markets.

If one adjusts Starlink’s EBIT for the Space segment’s launch costs of around $17.5 million per launch, Starlink only generates 11% operating margins (122 launches in 2025 x $17.5 million/launch = $2.14 billion). To be sure, if launch costs drop by 99% after Starship can actually fly, Starlink’s operating margins may indeed reach 21.3%, but this is hardly worth the lofty valuations being applied by Musk and his bankers to the IPO.

In the consumer segment of Starlink, rivals include HughesNet and Viasat (both of which have seen stagnant or declining revenue as Starlink steals their rural customer base) and Amazon's Project Kuiper (which is still in the deployment phase and not generating meaningful consumer revenue). In the enterprise/government segment, Eutelsat OneWeb is the primary competitor for now. Their growth has been modest, lacking the vertical integration and massive bandwidth scale that drove Starlink's $11.4 billion of revenues in 2025.

The projections for Starlink in SpaceX’s S-1 gloss over capacity exhaustion in high-density areas. Starlink is already facing bandwidth throttling in parts of the US. To grow revenue at the projected rate, they assume uninhibited international expansion, but they face severe regulatory bans or delays in critical high-population markets like India, China, and parts of Africa.

Why Starlink is a crappy business: With an active fleet of over 9,600 satellites and a lifespan of roughly 4 years per satellite, SpaceX must launch at least 2,000 satellites every single year just to replace de-orbiting hardware and maintain current coverage. To reach its 30,000+ goal by 2030, SpaceX must launch thousands more satellites per year. And this needs to be pulled off flawlessly every year in order to offset a constant 3 to 4-year decay rate of Starlink satellites.

With the huge earnings risks, the “pie-in-the-sky” goals, and stated corporate governance that says shareholders’ have zero say, it’s hard to see many long-term institutional investors buying the SpaceX IPO.

How I’ll Trade This Event (Not Investment Advice)

While I warned that the cancellation of the SpaceX IPO might lead to Tesla shares mooning last week (this is still a potential risk, which is detailed here), let’s now envision that the IPO roadshow starts on June 4th (as is already being heralded by brokers in Asia). Here’s what I’m looking for, unless the IPO gets canceled:

Read more

• xAI has become toxic to SpaceX’s IPO: The Paris Public Prosecutor’s Office raised their CSAM probe into xAI to a criminal investigation on May 7th, which has set off many alarm bells across Europe, especially among reinsurers of SpaceX’s Directors and Officers (D&O) insurance, as well as Errors and Omissions (E&O) insurance. Two of the world’s most powerful reinsurers are already re-evaluating the risks, based on a whistleblower’s alert (more details below), which means they’ve sent Advisory Notices to the IPO underwriters.

• SpaceX is more easy to merge with Tesla as a private company: There’s rife speculation about a Tesla-SpaceX merger, but this only makes sense if SpaceX remains private, as a merger after it goes public would be extremely complicated, costly, and time-consuming. If SpaceX is private, however, Tesla could easily do a “merger of equals” without too many hitches, as Musk has super majority voting power at SpaceX, and roughly the same at Tesla, given its supine Board of Directors. It would be almost as easy as SpaceX’s $250 billion acquisition of xAI on February 2nd, which is now the cause of the IPO becoming wobbly.

• A Tesla-SpaceX merger of equals is more lucrative than a SpaceX IPO: As detailed in my March 28th report on the SpaceX IPO (here), a $50 to $75 billion IPO of SpaceX would not only be the largest ever, but also suck a lot of capital out of the markets, especially from Tesla. Retail investors make up 40% of Tesla’s shareholder base, and many would likely swap their Tesla shares for SpaceX, which would weigh on Tesla’s stock price. If Tesla and SpaceX do a merger of equals, it’s not only a “win win” for both companies’ shareholders, but it would also give Musk around 32% voting rights of the merged entity—effectively veto power.

• Tesla would moon in the event of a merger with SpaceX: Assuming a merger of equals at a market cap of $1.5 trillion for both companies, the passive buying of Tesla would likely create history’s biggest buying squeeze (more details below). It would also likely create one of the biggest shorting opportunities after the passive buying is finished, as both Tesla and SpaceX are burning cash and are excessively over-valued.

Figure 1 below shows how quickly SpaceX’s valuation increased ever since Trump was elected President in November 2024. What’s even more mind-boggling is this:

• SpaceX’s valuation rose from $800 billion in December to $1.75 trillion in March, despite 2025 revenue growth of only 33%.

• February’s jump to a $1.25 trillion valuation from $800 billion in December was based on a SpaceX valuation of $1 trillion and xAI’s valuation of $250 billion.

• The final jump to a $1.75 trillion valuation came with the Confidential IPO Filing target with the SEC.

Figure 1: SpaceX Valuation Changes by Funding Rounds

Source: PitchBook and Forge Global.

Read more

Musk backpedaled on his robotaxi rollout plans on the Q1 earnings call, saying that robotaxi revenues won’t be “material” this year, and a more advanced version of Full Self-Driving (FSD) software was needed to improve safety. This is a sharp change after a decade of hyping autonomous vehicles and robotaxis.

Even Tesla’s Q1 2026 Shareholder Deck walked back its “Robotaxi Coverage” timeline, as is clear from the removal of their January plans of expanding robotaxis to 5 more cities (see Figures 1 and 2).

Tesla launched driverless robotaxi rides in Austin 4 months ago, given the lax laws there, but this could stop after May 28th, when stricter automated vehicle (AV) regulations go into effect (see Figure 3). Unless Tesla self-certifies FSD as a Level 4 automated driving system (ADS), running paid driverless rides there won’t be possible anymore. It’s doubtful that Musk will take such risks, given Tesla’s Austin fleet has already had 15 incidents in 7 months, including 2 injury crashes, despite having safety drivers in the front seat (see Figure 4).

Crash data filed by Tesla shows an alarming rise in accidents where the cause was FSD willfully violating traffic laws, like running red lights, stop signs, or driving in opposing lanes. After FSD crashes on non-highway roads jumped by 54% YoY to 515 cases in 2025, the total for January and February hit a record high of 108 crashes, an 80% YoY rise. Because of this, the National Highway Traffic Safety Administration (NHTSA) launched a deep probe into FSD last October, which unearthed 8,313 more potential FSD crashes for similar reasons.

NHTSA is probing whether FSD is a “scofflaw AI”: Unlike previous investigations, NHTSA is now asking Tesla to provide data on how FSD thinks when choosing to break traffic laws. Tesla has missed several deadlines to submit such data, which will likely weigh on any chances of it obtaining driverless ride-hailing permits in Texas or any other states. This should slowly weigh on the stock as most of its excessive valuation lies in hopes of Tesla scaling a disruptive robotaxi network from the 2H of 2026.

This report centers on a deep dive into the Level 2 ADAS crash data that Tesla submits to NHTSA, with its legal right to have NHTSA redact anything it deems as “confidential business information” (which is nearly everything that could help identify FSD incidents). As detailed below, this makes it extremely difficult to isolate FSD crashes, as they’re lumped into cases involving Autopilot and Enhanced Autopilot. After poring through several hundred thousand lines of data in these reports, my conclusion is that, as of mid-March 2026, FSD has an estimated 1,353 crashes since 2021, with 19 reported fatalities (which are under-reported), and 952 injury crashes (also under-reported). Tesla has systematically undermined NHTSA’s crash reporting process, which is why they’re under a separate NHTSA investigation (AQ 25002) into lack of data integrity.

Paid driverless rides in Tesla robotaxis in Texas will hit the same wall as was hit in Arizona and Nevada: Tesla was quick to obtain its commercial ride-hailing permits in Arizona and Nevada 8 months ago, but both states require a formal document from Tesla, which self-certifies FSD as being capable of Level 4 autonomy (FSD is only Level 2). This is why Tesla has yet to receive its driverless AV permit in both states, and when new AV laws become enforced in Texas from May 28th, the same should happen in Texas. If Tesla stops its paid unsupervised rides in Texas, it could weigh on Tesla shares.

The fact that Waymo has doubled its rides per week to 500,000 in under a year, while Uber has lined up 6 core partnerships for driverless ridesharing raises the question of whether Tesla will be relevant in this industry. For if it were valued as a normal carmaker, Tesla’s market cap would be no more than $100 billion ($27/share), implying that investors are awarding FSD and supervised robotaxis at $1.48 trillion, or $418/share. In light of the analysis of FSD crash data in this report, as well as the probe that it invited, it’s hard to see Tesla as a serious player in the driverless ride-hailing industry. In fact, it raises questions about whether Tesla needs to overhaul its entire FSD hardware and software stacks, which could put it further behind its rivals.

Because of the arduous process of isolating FSD-specific crashes in NHTSA’s Level 2 ADAS crash reports (which are heavily redacted on Tesla’s request), there is an Appendix at the end of this report that lists up my methodology. For the sake of accuracy and in the interest of time, I focused only on non-highway crashes, as this weeds out the once predominant highway crashes on Autopilot, which can’t be used to drive on city streets. Any suggestions or questions about process used for the non-highway crash counts in this report are welcome.

====================================================================

This report has 3 main sections and an Appendix following the Introduction, with 7 video clips of FSD in the wild, and 1 of Musk lying about FSD at January’s World Economic Forum, for comic relief. There are also 12 tables and charts worth looking at.

The “meat” of this report (the result of the long process of identifying undisclosed FSD incidents in NHTSA’s heavily redacted crash reports) is in Section 2, for those who want to skip ahead. The 3 sections and Appendix are labeled as follows:

SECTION 1: Tesla Has Lots to Hide in its Crash Reports

SECTION 2: How FSD Became Musk’s “Frankenstein”

SECTION 3: Tesla is Overwhelmed by 2 NHTSA FSD Probes

APPENDIX: Methodology For Identifying FSD Crashes in NHTSA’s Level 2 ADAS Reports

INTRODUCTION

Musk walked back Tesla’s robotaxi rollout plans: On the Q1 earnings call, Musk said that Tesla’s robotaxis would not be “material” to revenues this year, but will “probably be material in a significant way next year”. It was just last July when Musk said on the Q2 2025 earnings call, "I believe half of the population of the US will be covered by Tesla's Robotaxi by the end of the year”. This absurd proclamation was roughly the 10th time since 2016 that Musk has held out the “robotaxi” carrot in front of investors, only to fail in executing anything.

Tesla also altered its robotaxi expansion timeline: Not only did Musk tone down investors’ expectations on the Q1 call, but Tesla also watered down its timeline of expanding its robotaxis to 5 other cities outside of Texas. In January's Q4 2025 Shareholder Deck, Tesla had a table showing the timeline of its robotaxi expansion, which said it would roll out services to 5 other cities in “1H 2026”.

In the Q1 2026 Deck, however, the “1H 2026” timeline was replaced by a less specific schedule of “Preparations Underway” (see Figures 1 and 2 below), which is as meaningful as the the Tesla Roadster’s production status being labeled as “Design development”. This should’ve raised more concern amid shareholders, but Musk made sure that it marked the bottom of Tesla’s underperformance, as the stock is up by 28% since then.

Figure 1: Timeline of Tesla’s Robotaxi Ramp in January 2026

Source: Tesla Q4 2025 Shareholder Deck.

Figure 2: Robotaxi Ramp Timeline Removed in April 2026

Source: Tesla Q1 2026 Shareholder Deck.

But, this isn’t a simple delay. Rather, it’s a concession that the “Unsupervised” rollout is essentially trapped in a cage. You can’t scale a robotaxi fleet across 5 states when the regulator is still trying to figure out if your software is fundamentally incapable of obeying a red light or seeing in the sun (see details in Section 3 below).

Nevertheless, the pump goes on with driverless robotaxis being added to Tesla’s fleet in Texas: Despite this clear effort to tone down expectations among Tesla fans, who were vocal about their disappointment over the lack of driverless rides in Texas, Tesla proceeded to add 22 “unsupervised” robotaxis to its Texas since the Q1 earnings call. This seems more like an opportunity to do more testing while Tesla can, as stricter new AV laws become enforceable on May 28th (see my report on this here), and we’ll soon see if Tesla has registered for its AV permit to continue Unsupervised rides (see Figure 3).

Figure 3: Texas AV Permits Available Ahead of May 28 Deadline

Source: Texas DMV.

All Musk has to do is self-certify Tesla’s robotaxis as being capable of Level 4 autonomous driving, and the Texas DMV will grant Tesla the AV permit it needs. But this seems highly unlikely after seeing how bad Tesla’s FSD crash data is, and sheds light on why Musk has suddenly got “cold feet” about going driverless.

FSD crashes and fatalities more than doubled in 2025: The reams of data from Tesla’s Level 2 ADAS crash reports that were analyzed for this report help explain why Musk has become more cautious about going driverless: total FSD crashes on non-highway roads spiked by 54% YoY to 515 incidents in 2025, versus 335 in 2024, while FSD-related fatalities doubled to 6, and injury crashes spiked by 220% to 16. While Tesla reported that the FSD miles driven by its fleet only rose by 27% YoY in 2025 (despite 2 monthly free trials in 2025), note that this is much lower than the 220% growth in injury crashes and 100% rise in fatal crashes, which is why NHTSA is breathing down Tesla’s neck since October.

Tesla’s Austin robotaxi fleet is a prime example of why FSD is far from ready to go driverless in light of its track record. Despite having Tesla’s most advanced FSD hardware and software stacks (they’re said to be 6 months ahead of what customers can use), Tesla’s Austin robotaxi fleet racked up 15 crashes in 7 months through January 2026 (see Figure 3). While Tesla’s fleet is small and its miles traveled are still low, humans are 3.2x safer than Tesla’s “Supervised” Robotaxis, while Waymo is 21.2x safer on a miles/crash basis.

It should be noted that, despite a safety driver in the front passenger seat, there’ve been several crashes at slow speeds where the robotaxi plowed into fixed objects, yet the safety driver with his kill switch couldn’t prevent it. Two of the crashes had “minor injuries”, with one requiring hospitalization, which is horrible in the world of AVs. This too is a topic of concern at NHTSA, as FSD crashes in the past 2 years have come with much higher kinetics.

Figure 4: Tesla’s Robotaxi Crash Rate vs Waymo & Humans

Source: NHTSA, Tesla, & Waymo. Note that Tesla robotaxis have human safety drivers in the front passenger seat.

But the data that Tesla reports to NHTSA on crashes by Tesla owners using FSD is more unsettling. Because Tesla redacts most data that helps identify FSD-related crashes from those incidents involving Autopilot, I’ve focused only on non-highway crashes in this report, as only FSD can drive on city streets, and this is where most of the recent crashes have grown.

Video 1: Reuters’ Plight With Tesla’s Robotaxis in Dallas

Source: Reuters.

SECTION 1

Tesla Has Lots to Hide in its Crash Reports

Despite boasting about FSD “Supervised” (i.e., under human supervision) being 7x safer than human drivers without FSD, Tesla refuses to disclose any FSD crashes, which are lumped into one category that includes crashes on Autopilot, a cruise control system for highway driving.

Musk is so obsessively secretive about all things to do with FSD, that it’s starting to look like he’s hiding flaws, not business secrets. Here are some examples that boggle the mind, in light of all his hype over robotaxis:

• Tesla never followed through with its California AV permits: Despite having filed for an AV testing permit in California just weeks after Waymo in October 2014, Tesla never filed the required data to obtain the next of 7 permits needed to obtain a driverless ride-hailing permit in California (it ultimately took Waymo 9 years of testing to receive its ful license). Tesla refused to submit the annual FSD disengagement data required by the California DMV. When word of Musk taking his Robotaxi “dog & pony” show to the Bay Area spread to the CA DMV last September, it caused a lot of confusion, leading officials to publicly deny they’d given Tesla the proper permits to operate a robotaxi service in California.

• Tesla never filed a Voluntary Safety Self-Assessment with NHTSA: This is a 12-point report on any AV operator’s ADS, covering basic things like its ability to achieve a minimal risk condition (MRC) without human intervention if the system fails, as well as details of the ADS’s operational design domain (ODD), and Object and Event Detection and Response (OEDR) capabilities. FSD’s ability to achieve an MRC without human supervision is questionable, while its OEDR capabilities became the subject of an Engineering Analysis at NHTSA (see details in Section 3 below). Musk said that Tesla’s annual Safety Reports would suffice as a VSSA for NHTSA, and skipped filing the specific 12-point report. This is a major hurdle if Musk wants to continue operating his robotaxi fleet in Texas after May 28th, when the new AV laws go into effect, as the Texas DMV requires a copy of the AV operator’s VSSA that it filed with NHTSA.

On top of this, Tesla is hitting the same wall in Nevada and Arizona, where they started the application process 8 months ago, but are stalled when it comes to self-certifying FSD as an ADS capable of Level 4 autonomy. They obtained their commercial permit to operate a ride-hailing service (just like in Texas), but are taking longer than others to self-certify FSD.

This is a huge red flag, especially because of NHTSA’s Engineering Analysis on its vision-only system, which has led to fatalities due to the system not being able to properly see things in the sun’s glare or hazy conditions.

And this is likely why Tesla always exercises its rights to have NHTSA redact any of its crash reports that NHTSA publishes on a monthly basis, like the Level 2 ADAS report for consumer FSD crashes, and the Level 4 ADS report for its Austin robotaxi fleet. Figure 5 shows Tesla’s disclosures versus rivals in the Level 2 ADAS reports, while Figure 6 shows its redactions versus rivals in the ADS report.

Figure 5: Tesla Has Much to Hide in its FSD Crash Reports

Source: NHTSA.

Figure 6: Unlike Waymo, Tesla Hides Robotaxi Details

Source: NHTSA.

The redactions in these public reports from NHTSA led me to spend weeks and months, on top of years of analysis, buried in crash data, trying to figure out a “hack” for isolating FSD crashes from the previously more abundant Autopilot crashes.

It’s well known that Tesla redacts any details related to an incident’s “narrative”, and its operational design domain (ODD). But I realized that the other fields that Tesla didn’t redact could help identify FSD crashes, and here’s how it went:

• City streets are the key: Since Autopilot can only be used on highways, I got excited after discovering these 2 categories in the old version of the Level 2 ADAS reports: “Roadway Type” and “Posted Speed Limit (MPH)”. Because Autopilot crashes are only on highways, I sought evidence of FSD crashes on non-highway roads, as that’s where the more recent versions of FSD have experienced most of their troubles. Unfortunately, 30% of Roadway Type entries are “Unknown”, while 17% of Posted Speed Limits were left blank.

• Injury and fatal crashes provide more solid comps: Due to all the redactions and omissions, it became clear that getting a clean track record of FSD incidents on non-highway roads would be difficult. This led to my decision of simply comparing injury and fatal crashes to luxury car crashes, given the driver profile was more similar to Tesla drivers (wealthier, older, and more cautious than the younger profile in the US national average). In the category of “Highest Injury Severity Alleged”, one can find everything from property damages, and minor injuries, to serious injuries and fatalities. Unfortunately, 89% of this category had “Unknown” entries (95% in the new format of reporting).

• Injury crashes can be isolated via airbag and tow-away data: Because Tesla is clearly hiding any crash data involving serious injuries and deaths, I thought that any incident where (a) an airbag was deployed + (b) the Subject Vehicle (Tesla) was towed away, would be a sure shot for at least an injury incident. This could be used to compare to miles per serious injuries or worse for luxury car drivers. Tesla’s airbag deployment data was robust, with over 95% of crashes showing airbags being deployed, which is great. But, since many of these on city streets did not lead to injuries, I needed the “Vehicle Towed?” category to confirm a serious injury crash. Unfortunately, 81% of this category was riddled with the “Unknown” entries.

• Musk forces through reporting changes at NHTSA after joining the Trump Administration: As early as December 2024, Musk had an automotive “First 100 Days” plan mapped out for the Trump Administration’s transition team, which led to last June’s change in Level 2 ADAS reporting rules (see headlines at the time in Figure 7). While the changes were made in order to “streamline” crash reports, the reduction of reporting categories from 137 to 116 really made things hard. I lost some favorites in the old format like “Posted Speed Limit (MPH)”, which was so helpful, as anything under 50 MPH could generally be considered an FSD crash on city streets. The field with the mileage of each car involved in the crash was also removed, which wasn’t huge, but stopped my analysis of how many miles each car had driven before its crash (the average was 32,252 miles, which is extremely short). And details on the “counterparty’s vehicle” in each crash were also removed, which makes it hard to assign blame in each incident. I’m convinced that all of these changes were made in order to hide Tesla FSD’s growing crash count, and the rising G-forces in each collision.

Figure 7: Musk Moved Quickly to Change Crash Reporting Rules

Source: Reuters.

• New reporting rules led to fewer entries by Tesla: Despite the 15% cut in reporting categories, it seems like Tesla simply used the opportunity to fill as many categories with “Unknown” as possible, as if to avoid reporting the details each category required for legal reasons. Figure 8 below shows how much the % of “Unknown” designations increased in the New Format versus the Old Format, and it’s quite telling: Injury and fatal crashes are now all but undisclosed, as 95.3% of the entries are now “Unknown” since last June, while crash details in “Subject Vehicle Pre-Crash Movement” and “Crash With” saw huge jumps in “Unknown” weightings. The sudden sensitivity to crash details is undoubtedly due to Musk having pushed for FSD to become more “human-like”, by making it more assertive, which led to higher-kinetic crashes on city streets. With these changes in reporting, Musk completely sealed off any chances of getting a good read on FSD crashes, which led me to hit a dead end: without being able to even get a proper miles/injury crash reading, there was no chance of doing a proper analysis that would include incidents per million miles, etc. This led me to simplify the search criteria, which is detailed in the Appendix below.

Figure 8: Weight of “Unknown” Entries by Crash Category

Source: NHTSA & MH research.

Video 2: FSD Abuse: 87-Year-Old Behind the Wheel

Source: Twitter.

Video 3: FSD Abuse: Typing on Laptop While Driving

Source: Twitter.

SECTION 2

How FSD Became Musk’s “Frankenstein”

The bulk of crashes reported by Tesla to NHTSA through 2022 were on the highway on Autopilot, a cruise control system that can’t be used in the city. But ever since FSD v12 was released in late 2023, the weight of FSD incidents reported by Tesla, mostly on city roads, increased in its crash reports (see Figure 9).

Version 12, or v12, was the first time that FSD began operating on an an end-to-end neural network, or driving on what it had learned by watching how the best drivers in FSD beta drove, rather than following “if/then” rules. This created a buzz among fans, as FSD suddenly drove more smoothly.

Musk’s Push For More “Human-Like” Driving Causes Problems

However, problems began in early 2022, when FSD had a recall issued by NTHSA because they found that Tesla had programmed it to do “rolling stops” at 4-way stop signs, or slowing down to only 9 km/hour (5.6 mph) while “rolling” through the intersection, rather than coming to a complete stop.

Tesla fixed the issue via an over-the-air update, but this marked the start of a struggle that continues today between Musk trying to get FSD to drive more like humans, which requires FSD to become increasingly “aggressive”. Tesla was concerned about FSD being "excessively hesitant", as in clogging up roads by deliberating what to do at 4-way stop signs, or failing to merge into heavy traffic because it was waiting for a "perfect" gap that never came.

Musk didn’t like this and made FSD increasingly aggressive as such hesitation was causing FSD beta testers to intervene by taking over, which drove down the "intervention-free" miles that Musk needed to prove his Robotaxi business case.

FSD only became more aggressive since then (see Videos 3 - 5 below), and the number of crashes on city roads began surging from 2024 (see Figures 9 and 10), with the wide release of FSD v12 and Musk’s efforts to lure more users via 1-month free trials and forced test-drives when receiving delivery of a new Tesla.

Video 3: FSD Goes Straight in Right-Turn Lane

Source: Twitter.

Video 4: FSD Drives into Opposite Lane

Source: Twitter.

Video 5: FSD Tries to Enter a “Do Not Enter” Path

Source: Twitter.

Figure 9: FSD City Driving Crashes (Estimates)

Source: NHTSA & MH research.

The numbers in the graph in Figure 9 and table in Figure 10 are my estimates of FSD crashes that Tesla reported to NHTSA, based on NHTSA’s monthly Level 2 ADAS crash reports, in which most key data is heavily redacted by Tesla’s choice.

While the details of my search criteria are explained in the Appendix at the end of this report, here are the main things I used to isolate only FSD crashes from others that occured on Autopilot or Enhanced Autopilot:

• Roadway Type: Used all types except highways/freeways.

• Posted Speed Limit: This highly useful category for identifying FSD crashes was removed by NHTSA in mid-June 2025, after Musk got the Trump Administration to change the overall reporting format for Level 2 ADAS crashes. I screened for any incidents where the posted speed limit was at 50 mph or lower. Some crashes happened while FSD was driving on city streets where the posted speed limit was 60 mph.

• Injury and fatal crashes: As explained above, 95% of entries in this field are listed by Tesla as “Unknown”, and hardly ever updated. After NHTSA launched a probe into Tesla’s dodgy reporting practices last August, Tesla has begun updating this category with fatalities and serious injuries, but the process is slow.

• ODI complaints: The Office of Defects Investigation (ODI) at NHTSA mainly analyzes safety-related defects in cars, but also fields complaints drivers. The complaints from drivers over near misses on FSD are used in the table below, simply as a gauge to track reported crashes for similar reasons. Note that since the more aggressive FSD v14 was released in October 2025, complaints to the ODI have surged, with 2026 YTD incidents already surpassing all of 2024. Also note that the more “emotional” complaints were removed (e.g., those with unrealistic claims, such as “90 fatalities” in an FSD near miss incident).

Figure 10: FSD City Crashes & Serious Incidents (Estimates)

Source: NHTSA SGO Level 2 ADAS reports & MH research.

Video 6: FSD Tries to Drive into a Lake

Source: The Dawn Project via Twitter.

Injury and Fatal Crashes Grow More than FSD Miles Driven

While the number of FSD crashes identified in Figure 10 spiked by 54% last year, injuries rose by an even higher 220%, while whatever deaths Tesla chose to report doubled.

What’s alarming is that total FSD miles driven in 2025 only grew by 27% to 4.25 billion miles (6.84 billion km), which not only points to lower interest in FSD after a 167% YoY surge in 2024, but also a higher crash rate per miles driven.

It’s crucial to put the FSD crashes into the context of “incidents per FSD miles driven”, which is shown in Figure 11, however it’s important to note the following:

• The miles-per-crash numbers in the US national average and the luxury car average are based on all-road incidents, including highway driving, which is much safer than city driving. This makes the comparisons with FSD miles/crash highly conservative, as I’m using only FSD non-highway crashes and miles driven. If the same city miles/crash data were available for the US national average and luxury cars, FSD would likely pale in comparison, from a safety perspective.

• FSD with human supervision is only 1.3x safer: In terms of miles/crash and miles/injury crash, FSD is only 1.33x safer than the luxury car average (more pertinent than the US national average from a demographic and safety equipment perspective), while FSD is 1.4x safer on a miles/fatal crash basis. Again, the FSD data is only non-highway miles, which is being compared to all-road incidents, including much safer highway miles driven.

• FSD crash data through the full month of February 2026 is matched by cumulative city miles driven by FSD through the end of February in Figure 11 below.

Figure 11: FSD Supervised is Not Much Safer Than Human Drivers

Source: NHTSA SGO Level 2 ADAS reports, NHTSA 2024 FARS Annual Report File, FHWA 2025 Statistical Projections, IIHS-HLDI 2025 Insurance Loss Report (Luxury Passenger Vehicle Segment), Tesla, and MH research. Note: FSD data is non-highway data, while the US & Luxury Car averages are all-road data.

The fact that, even with human supervision, FSD is not as safe as human drivers, doesn’t bode well for the outlook of Tesla’s “robotaxi” expansion.

Video 7: Musk Stutters While Lying About FSD Safety at WEF

Source: Twitter.

SECTION 3

Tesla is Overwhelmed by 2 NHTSA FSD Probes

NHTSA issued a recall of FSD in 2022 due to Tesla having programmed it to do “rolling stops” at intersections with 4-way stop signs, in an effort to reduce the rate of intervention by drivers, which made FSD look less safe. Musk has thumbed his nose at NHTSA in the past and he likely did the same after this recall, as FSD has only become more aggressive, and NHTSA has received a flood of complaints about near misses due to FSD ignoring traffic rules.

Tesla’s “robotaxi” launch in Austin is said to have piqued NHTSA’s interest in FSD’s tendency to willfully break violate traffic laws, which had led to mounting crash incidents. Out of the 1,353 FSD non-highway crashes I’ve identified from NHTSA’s crash reports, 25% of them occurred at intersections, and 20% of theme where while FSD was making “complex maneuvers”.

Video 8 was said to have gone viral and caught NHTSA’s attention, according to news reports, and NHTSA was said to have requested a test ride in a robotaxi shortly after seeing it.

Video 8: This Viral Clip Caught NHTSA’s Attention in Austin

Source: Twitter.

NHTSA Probes the Machine’s Intent

With the launch of NHTSA’s Preliminary Evaluation (PE 25012) of FSD’s frequent violation of traffic laws last October, and the Engineering Analysis (EA 26002) of FSD’s inability to detect objects in hazy conditions in March, Musk is coming closer to meeting with karma after years of thumbing his nose at NHTSA.

These 2 investigations delve into both the software and hardware stacks of FSD, and while the skeptics may see this as a “nothing burger” (NHTSA has never issued a recall of Tesla’s entire FSD system up to now), their mere existence will make it all but impossible for Tesla to obtain driverless ride-hailing permits in most states.

It’s hard to ignore the irony of the fact that these two momentous probes of Tesla’s FSD—the key factor behind Tesla’s $1.5 trillion market cap premium as a small carmaker—are being conducted by NHTSA’s Office of Defects Investigation (ODI), which Musk gutted during his tenure at DOGE last year.

He specifically targeted staff cuts among data analysts and software engineers within the ODI who were in charge of autonomous vehicle safety oversight. Musk also reduced headcount in the Vehicle Control Division, which is in charge of verifying Tesla's telemetry data, leading to the glacial pace of the current audit.

Last December, the ODI sent Tesla a list of 11 questions on how FSD thinks when it willfully violates traffic laws, after fielding 80 complaints of near misses on city streets by FSD users (see ODI’s Information Request here). To get a sense of how deep the ODI is probing, see Figure 12, which shows question 8, the toughest of 11 in their Information Request.

In its exchanges with the ODI regarding these questions, Tesla whined about the ODI’s Information Request being “unprecedented in scope” and “unduly burdensome”.

Tesla has asked for and received 4 deadline extensions, pushing the final verification dates from early Q1 into the summer of 2026. Despite these extensions, Tesla has likely missed at least 2 deadlines regarding the submission of high-fidelity video logs, as well as a response to Question 8, the toughest one, which deals with FSD’s decision-making logic (see Figure 12).

Despite Musk’s claim that "Tesla is by far the best in the world at real-world AI", they can’t seem to provide timely replies to NHTSA’s questions on how its AI thinks. In its probe on why FSD’s vision-only system often goes blind when directly facing the sun’s glare, or driving in foggy or dusty conditions, Tesla also slow-walked their answers to NHTSA’s questions. This led to the agency elevating their Preliminary Evaluation of FSD’s “vision degradation” to an Engineering Analysis, the last phase in deciding whether a recall is needed.

What is FSD Thinking? Tesla Struggles to Reply

The most critical threat to Tesla’s stated goal of expanding its robotaxi network is PE 25012, NHTSA’s probe into FSD’s decision-making logic when it chooses to break traffic laws. Unlike previous inquiries that focused on hardware performance or driver monitoring, this probe dives deep into how and why the software thinks the way it does. NHTSA is specifically investigating why FSD appears to be willfully running red lights, ignoring stop signs, and veering into opposite lanes of traffic (see examples above in Videos 3 -5).

In their 13-page Information Request (IR) sent to Tesla last December (here), NHTSA explicitly demands to know if FSD can “accurately detect and appropriately respond to traffic signals”. Tesla’s April 2026 response failed to resolve these concerns. If the data for 2026 shows the same violations, it indicates that Tesla’s “end-to-end” neural net is failing to prioritize legal compliance over behavioral imitation.

Figure 12: NHTSA’s Toughest Question for Tesla in its FSD Probe

Source: NHTSA Information Request ID PE25012-01.

If Tesla admits that the neural net “prefers” human-like flow over strict legality to avoid being “boring” or “hesitant,” they’re effectively admitting that FSD is a scofflaw. If they claim the law is always the top priority, NHTSA can point to the hundreds of video clips (like those above) showing the car willfully breaking traffic laws.

This ties directly into the 416 city-street crashes identified in this report. In many of these cases, the telemetry suggests the system didn’t just fail to see an object, but more importantly, it appears to have chosen a path that was illegal and dangerous.

The Probe Finds 8,313 Unreported FSD Incidents

After launching its probe last October, NHTSA wasn’t receiving the answers it asked about why FSD was persistently violating traffic laws, so they told Tesla to do keyword searches of their databases, including phrases like “ran a red light”, “drove the wrong way”, etc.

This led to the unearthing of 8,313 unreported FSD incidents, which is immense: if half of these crashes are added to my estimated 1,353 city crashes, FSD’s miles per crash drops to 537,000 miles from 2.3 million miles, and would render FSD as safe as human drivers, but raise luxury drivers to 1.3x safer than FSD.

Tesla’s slow-walking of its answers to NHTSA’s questions is likely directly linked to Musk choosing to tone down investors’ hopes of a rapid ramp-up of robotaxis across multiple states, as well as Tesla unlocking consumer FSD to “Unsupervised” status, for participation in its Robotaxi network.

This is how much Tesla is struggling to reply to NHTSA’s questions, which raises questions of whether they truly have a veritable AI in their own systems:

• Slow verification: To date, Tesla has submitted roughly 960 reports.

• Low-quality replies: Most of these remain “unverified” in the SGO database, yet the telemetry and narrative evidence strongly suggest FSD was at fault.

• Robotaxis could be delayed for 6 years: At the current pace of 100 reports per month, it will take Tesla nearly 6 years to finish the audit. By cutting the internal audit teams responsible for this data cleanup in early 2026, Tesla has turned a regulatory deadline into a perpetual backlog. NHTSA is no longer just “irritated”; they’re watching a company prioritize expansion while effectively stonewalling the data surrounding its past failures, and this will directly block the rollout of Tesla’s driverless robotaxis.

FSD’s Vision Problem: Engineering Analysis 26002

The March 2026 elevation of the “Vision-only” hardware probe into an Engineering Analysis (the final stage of an investigation before recall decisions are made) is a multi-billion-dollar threat to Tesla. This investigation focuses on whether Tesla’s camera-only hardware system can handle complex lighting impairments—such as low-angle sun glare, heavy shadows, or airborne dust—which we’ve seen pop up in dozens of the crashes.

NHTSA’s review of the Austin robotaxi incidents showed that FSD frequently failed to detect roadway conditions that would cause it to be blinded until immediately before a crash. Tesla admitted that its updated detection system would have only potentially affected 3 out of 9 identified crashes, meaning the update would not have helped in the majority of cases.

• Prospects of costly retrofits: If NHTSA determines the current cameras are physically incapable of resolving these edge cases, Tesla could be forced into a massive retrofit program, which would cost over $4 billion just for its US vehicle fleet.

• FSD is useless without proper vision capabilities: In my analysis of the FSD crash data, there were incidents where the vehicle didn’t even attempt to brake. This isn’t a software bug; it’s a sensory failure. If the “eyes” of the car are blind in certain lights, the “brain” cannot prevent the collision.

After writing this report, I’ve concluded that Musk and his robotaxi team realize that they’re dealing with bugs in the FSD software that could require a total revamping of both the hardware and software stacks.

This would not only cost tens of billions of dollars in retrofitting customers’ cars, but could bankrupt Tesla in legal liabilities: investors could sue for losing money on Tesla’s stock, while customers could sue for having bought a Tesla on Musk’s promise of being able to make money from having their cars participate in Tesla’s robotaxi fleet.

This leads me to believe that the SpaceX IPO may be skipped in order for a quick and easy merger with Tesla, which would be a win-win situation for both Tesla and SpaceX investors, rather than a loss for Tesla investors and a win for SpaceX investors. Look forward for a report on this by Friday.

APPENDIX

Methodology For Identifying FSD Crashes in NHTSA’s Level 2 ADAS Reports

NHTSA’s data is set up in an archaic, hard-to-use style, which is why many analysts—including yours truly—give up on analyzing the monthly crash reports, especially because NHTSA redacts anything Tesla asks it to before publishing its reports (see below).

Tesla has 3 different Level 2 ADAS systems: Autopilot (AP), Enhanced Autopilot (EAP), and Full Self-Driving (FSD). AP and EAP are used for highway driving, with AP only capable of driving straight forward on highways with good markings, while EAP allows for lane changes. Neither can drive on city roads. FSD can drive on both city roads and highways.

While I spent 2 weeks immersed in crash reports between 2021 and March 2026, the final search criteria used in this report were very detailed at first, until things hit a wall of inconsistencies, due to Tesla’s numerous redactions. The end-product is a simplified list of search criteria—which is likely as close as one can get with the limited disclosures—which is better, and less time-consuming.

For reference, my original search fields are listed below those used below:

Previous Format Search Criteria (May 2021 - June 15, 2025)

1. Non-highway crashes excluding “Unknown” incidents: 661.

2. “Unknown” crashes where posted speed limit was 50 MPH or lower: 183.

3. “Unknown” crashes with both airbags deployed and tow-aways: 41.

New Format Search Criteria (June 16, 2025 - March 15, 2026)

1. Non-highway crashes excluding “Unknown” incidents: 218.

2. “Unknown” crashes under 50 MPH with airbags deployed: 222.

3. “Unknown crashes” under 50 MPH with airbags deployed and tow-aways: 23.

It should be noted that the above search criteria is very conservative for the following reasons:

• Some FSD crashes on “City Streets” occured with the posted speed limit of 60 MPH (97 KPH).

• The 3.23M miles/injury crash estimate for Tesla is a conservative estimate because it assumes every single airbag deployment resulted in an injury, whereas the human benchmarks rely on police-reported injuries, which are crowded with fender benders, and other lower-kinetic incidents.

Other search criteria considered, but not used due to a lack of consistency linked to Tesla’s redactions:

1. Any category that includes the words “See Narrative”, such as the “Weather - Unk - See Narrative”, as well as “Roadway-Other-See Narrative”, where Tesla sometimes answers “Y” for “Yes”.

2. Any “Y” for “Yes” reply from Tesla in the category of “Source - Complaint/Claim”, as this is the second most used source of crashes after Telematics.

3. Any “Y” for “Yes” reply from Tesla in the category of “Weather - Fog/Smoke/Haze”, given Tesla’s vision degradation problems.

4. Any “Y” for “Yes” reply from Tesla in the category of “Weather - Rain”, given that the Austin “unsupervised” robotaxis don’t operate on rainy days.

5. Any “Y” for “Yes” reply from Tesla in the category of “Weather - Dust Storm”, given Tesla’s vision degradation problems.

6. Any “Y” for “Yes” reply from Tesla in the category of “CP Contact Area - Rear Left”

7. Any “Y” for “Yes” reply from Tesla in the category of “CP Contact Area - Rear”

8. Any “Y” for “Yes” reply from Tesla in the category of “CP Contact Area - Rear Right”

9. Any “Y” for “Yes” reply from Tesla in the category of “Data Availability - Complaints”

10. Any reply from Tesla in any category where they say, “Other, see Narrative”, as that’s usually a reply to hide that it’s FSD.

11. We should also expand our FSD filters to include any Tesla Level 2 ADAS crash report that has the following characteristics in the “Old” format (Excel file attached):

12. Any “Y” for “Yes” reply from Tesla in the category of “Source - Complaint/Claim”, as this is the second most used source of crashes after Telematics.

13. Any “Y” for “Yes” reply from Tesla in the category of “Weather - Fog/Smoke” given Tesla’s vision degradation problems.

14. Any “Y” for “Yes” reply from Tesla in the category of “Weather - Rain”, given that the Austin “unsupervised” robotaxis don’t operate on rainy days.

15. Any “Y” for “Yes” reply from Tesla in the category of “Data Availability - Complaints”

16. Any reply from Tesla in any category where they say, “Other, see Narrative”, as that’s usually a reply to hide that it’s FSD.

17. Any incident (mostly on streets and intersections) where the Tesla was making a left turn while the crash partner was proceeding straight, especially if the Tesla’s airbag was deployed.

18. Any incident where Tesla replies “Unknown” to key crash detail categories like Roadway Type (street, intersection, highway, etc.), Crash With (what the Tesla crashed into), Crash Partner Pre-Crash Movement, Subject Vehicle Pre-Crash Movement, while replying “Y” for “Yes” in categories asking if data was available the car’s Event Data Recorder (EDR), telematics, and videos. This is usually a blanket reply in FSD crashes.

NOTE: NOTHING IN THIS REPORT IS INVESTMENT ADVICE.

As I mentioned in my Q1 preview report on Monday, this quarter’s earnings call will be more important than the results themselves, and this was exactly the case last night: Tesla’s headline Q1 EPS of $41 was a 24% beat, which sent shares up by 4.2% to $404 in after-hours trading, before dropping by 5.4% to a low of $385 as some very unsavory details were announced.

The bad news—as expected—was that Musk had to tone down his hype on Tesla’s robotaxi expansion, as mounting FSD crashes among Tesla owners are more dangerous, and the Austin robotaxi fleet already had 2 injury crashes (see details in this report).

And to top it off, Musk clearly stated that Tesla cars with Hardware 3 (HW3) computers would not be capable of fully autonomous driving, and that Tesla would need to set up “micro factories” in large cities to carry out the retrofits (no timeline or cost estimates were given, but see estimates below).

And while many fans were relieved that Tesla managed to generate positive free cash flow (FCF) of $1.4 billion in Q1, the stock sunk even lower after Tesla’s CFO warned that Tesla would burn cash from Q2 onward as it raised its 2026 capex budget by 20% to $25 billion (2.3x GM’s 2026 capex plans of $11 billion).

This report focuses mostly on the robotaxi issue and the estimated costs of upgrading customers’ Hardware 3 FSD computers to Hardware 4, but below are the key earnings points:

• Q1 EPS beat of 24% was a 15% miss when one-offs are excluded: There were a lot of one-off benefits in Tesla’s Q1 earnings, including $200 million of forex tailwinds, $230 million of one-off warranty true-down (likely related to the discontinued Models S & X), over $250 million in tariff benefits in the Energy division, and an unquantified amount of “tariff relief”. Figure 1 below shows the adjustments, amount to an adjusted EBIT of $251 million (versus the $941 million result), and a loss of $129 million excluding regulatory credits.

• Vehicle prices rose sequentially despite heavy discounts: While a weaker US dollar versus the Chinese yuan and other currencies like the Norwegian krone led to higher ASPs on average, it’s hard to believe that Tesla saw as much as a 5% rise in ASPs, even after the currency impact is excluded. Price cuts in Germany and Norway were huge, and most regions saw 0% loan programs being offered. Cost per unit rose by 3.4% sequentially, which may be a sign of higher lithium prices starting to seep in (lithium prices have risen by 40% from their Q2 2025 lows).

• Tesla Energy saw high profitability, which should drop from Q2: Even after excluding the one-off tariff benefit of “over $250 million” in Tesla’s Energy division, gross margins were still high at 29% despite a 38% decline in GWh deployed in Q1. Moreover, the revenues per GWh deployed was up by 1% QoQ, despite heavy price competition in the energy storage space (see details in my Q1 preview here). Tesla’s CFO warned that margin compression is expected from Q2 onwards, due to higher tariffs (LFP cells from China will see a 28.4% tariff this year, up from 10.9% in 2025).

• Inventory increase was mostly due to Tesla’s Energy business: While Tesla produced 50,363 more cars than it sold in Q1, most of the $2.26 billion drag on cash came from Tesla Energy, which saw a 31% sequential rise in inventory (Automotive Inventory only rose by 12% QoQ). This points to a more robust Q2 deployment of energy storage systems, as demand is strong, but Tesla’s vehicle inventory of 128,888 units is a sign that more price cuts and incentives will be needed in Q2, while lower output will lead to less fixed cost absorption.

• Q1 capex was only 10% of full-year guidance: Even under its original target of 2026 capex “in excess of $20 billion”, Q1 capex of only $2.5 billion was low, but in light of its new plans for $25 billion in full-year capex, Tesla must spend $7.5 billion per quarter from Q2 through Q4. Given dwindling profis (Q1 GAAP net profit was roughly flat despite over $500 million of one-off tailwinds and 16% YoY revenue growth). Cash burn in 2026 should be $17.6 billion by my estimates, or 40% of Tesla’s Q1 cash.

• Regulatory credits set to plunge from Q2: The Trump Administration canceled the ZEV credit system in the US last year, and nullified the greenhouse gas (GHG) credit system in February. As explained in my March 31st report (here) on this issue, GHG credits made up the bulk of Tesla’s huge regulatory credits in 2024 (39% of EBIT). Based on Tesla’s Q1 10-Q, they state that obligations over 1 year have dropped to only $303 million in Q1 (from $841 million in Q4 2025, and $4.6 billion in Q1 2025), of which Tesla only plans to recognize $198 million over the next 12 months. While this implies only $50 million per quarter from here, it should be noted that what Tesla states it expects to recognize is usually lower than what is booked in the following quarters. However, with the bulk of its credits gone, the contribution from regulatory credits should be much lower, at around $200 million per quarter, at best.

• Debt in China rises and a new $1.5 billion credit facility was formed: Tesla drew $1.5 billion of debt from its China Working Capital facility (an increase of 35% QoQ to $5.8 billion), and also opened a new $1.5 billion “Warehouse Agreement” with its banking syndicate—something that hasn’t been seen since 2020. While Tesla has yet to draw from this credit facility, it’s another sign that Tesla is now on the defence, as it expects huge cash burn.

• None of Tesla’s $25 billion of capex is for new EVs: Tesla’s CFO said that the 2026 budget for $25 billion of capex is needed in light of Tesla building 6 factories this year, without giving a breakdown. Most of this is going into AI infrastructure and Optimus, as the other factories are already built or close to completion. The fact that none of this is going to new EV models—Musk said that the Cybercab would suffice—it points to a “make or break” strategy that’s highly risky. Tesla is essentially spending over 2.3x 2026 capex plans of $11 billion at GM, which sells over 3.8x more cars than Tesla does. Investing in AI infrastructure and new humanoid bot lines will do nothing to stop the need to cut prices in order to move the metal, so the real question is how long can Tesla last without new models? Because, without a functional automated driving system (ADS), Tesla won’t be generating any cash from its robotaxi ambitions.

====================================================================

Figure 1 shows Tesla’s earnings adjusted for one-off benefits Tesla detailed in its earnings release, as well as on its earnings call. The adjustments made in Figure 1 add up to $690 million and include the following:

Read more

• Tesla Q1 EPS should be in line with consensus: The Tesla-compiled consensus on April 17th sees Q1 2026 adjusted EPS of $0.33 (+24% YoY), on 11% YoY revenue growth, and 36% YoY EBIT growth, for an EBIT margin of 2.5%. This is the first time my estimates are above consensus, with only 8% revenue growth, but a 61% rise in EBIT, with an EBIT margin of 3.1%. My EPS estimate is the same as consensus at $0.33, although my cash-burn estimate is 24% higher at -$1.96 billion (see Figure 1 below).

• The Street’s full-year estimates are too high: The Tesla-compiled consensus sees 2026 adjusted EPS of $1.87 (+13% YoY), on 5% revenue growth and 3% EBIT growth, with an EBIT margin of 4.5%. FCF estimates are at -$6.58 billion. My estimates see 2026 revenues dropping by 6% YoY, with an operating loss of $1.18 billion, and cash burn of $11.92 billion.

• EV subsidy programs boosted Q1 deliveries massively: In Thailand, where EV subsidies were set to get cut from February, a rush in demand led to Tesla registering (but not selling) 5,800 cars, which is more than the 5,049 cars Tesla sold in Thailand for all of 2025. In France, Germany, and Spain, new EV incentives worth $4,722 on average helped spike Tesla’s sales in those markets. Tesla still offered hefty incentives via aggressive low-interest loan programs in many of such countries.

• Price cuts only in Korea and Germany; 0%-interest loans elsewhere: While Tesla only cut prices in Germany and Korea (by $3,493 on the Model Y in Germany, and by nearly $9,000 on average in Korea), its widespread use of 0% loan programs in most other regions should’ve slashed profit margins in Q1. Without any new models in its pipeline, this will continue, unless Tesla closes a factory to reduce the burden of having to move the metal to keep its factories humming.

• Tesla Energy is in for a bad year after Q1: The tariff on imported LFP cells from China, used in Tesla’s highly popular Megapack energy storage systems (ESS), went from 10.9% in 2025 to 28.4% as of January 1st this year. While Q1 earnings won’t be impacted due to Tesla reportedly having hoarded LFP cells from China in late 2025 ahead of the tariff hike, Q2 onwards will be tough, with only 18% gross margins versus last year’s record 28.9% margin.

• Musk has lots to answer for on the earnings call: This Q1’s earnings call is more important than those in the past: Tesla is still using safety drivers in its Austin fleet, despite Musk having promised 5 times last year that Tesla’s Austin robotaxis would be driverless by the end of 2025. Moreover, if FSD isn’t safe enough for paid driverless rides, what will happen to the Cybercab, which was scheduled for mass production start this month? This 2-seat EV has no steering wheel or gas/brake pedals, which means no one can drive the car. After 3 years of development, this will be a huge loss even if they connect the steering wheel and gas/brake pedals, as Americans only buy 2-seaters if they’re sports cars.

• Watch for any excuses for Tesla robotaxis not expanding more: While Tesla did another PR stunt over the weekend by showing footage of driverless Tesla robotaxis having expanded to Dallas and Houston, this won’t be possible after May 28th, when the Texas DMV starts enforcing stricter AV laws that became law last year (see my note on this from yesterday here). Even in very lax states like Arizona and Nevada, Tesla has yet to launch driverless services due to the need to self-certify FSD as a Level 4 automated driving system (ADS). Aside from delayed rollouts in most cities, it will be key to hear any excuses for not going driverless in any of these new areas of operation.

• The “robotaxi” theme makes up $1.3 trillion of Tesla’s market cap: On a sum-of-the-parts basis (see details near the end of this recent report), Tesla should only be worth $26 per share, which implies that 94% of its $401 share price, or $1.3 trillion of market cap, is attributed to Tesla’s non-existent robotaxi business. If Tesla has no paid driverless services by the end of Q2, this massive valuation premium should start falling apart. Despite the hype by Musk and his Tesla influencers, Tesla’s FSD is regressing in capabilities, with accidents caused by FSD willfully deciding to break traffic laws (see details in my March deep dive here).

Valuations may start to matter: Many commentators have begun quoting Tesla’s nose-bleed valuations this year (214x consensus EPS estimates of $1.87), most likely because it’s seeing revenue declines, along with deteriorating earnings. Tesla’s current market cap of $1.4 trillion is 36% higher than the top-15 carmakers in the world, which sell 98% more vehicles than Tesla does. If the robotaxi dog and pony show falters this year—which is my base case—this valuation premium that has made Tesla more richly valued than even the AI hyperscalers, should collapse.

Read more

Tesla posted a video of a driverless robotaxi driving in Dallas and Houston to announce their “robotaxi” services now operate in 4 cities, after having launched in Austin and the Bay Area last year (see Video 1).

A few influencers also uploaded their experiences, one of which had lots of mishaps (see Videos 2 and 3).

The fact that this comes just days …

Read more

The war in Iran has led to the largest disruption to oil supply in history, dwarfing the oil crises of the 1970s, and, for the first time, also impacting the supply of LNG. Despite the implications of global stagflation, the US equity market is utterly unfazed and is now slightly up by 0.1% since the war began on February 28th.

The same is largely true for many auto stocks with large earnings exposure to the US market, which is seeing the highest spike in gas prices in the world, after having removed all incentives for EVs over the past 6 months. This means that car-buying decisions in the US, where affordability has crimped demand amid record-high new car prices of $50,000, should increasingly focus on fuel efficiency.

And this is where it’s worth looking at who the winners and losers could be, as you can likely catch fish without even trying. If the oil price remains above $100 for a prolonged period of time, it could squash US car demand. Here are the main points:

• The US price of gasoline is up 38% in 6 weeks: The US has seen the sharpest spike in the price of gas, which is up by 380% to $4.11/gallon ($1.09/liter) since the war in Iran began. For a country where people drive an average of 13,476 miles (21,687 km) per year—twice that of their European and Asian counterparts—this will be problematic and should force a change in car-buying behavior like it did in 2008.

• The Detroit 3 have the lowest fleet mileage and face the biggest risks: GM, Ford, and the Chrysler unit of Stellantis have the lowest fleet mileage in the industry, with their top-selling pickup models getting only 20 miles per gallon (5.3 km/liter) versus their rivals’ average of 28 mpg (7.3 km/liter). The Detroit 3 face the biggest drop off in US sales if gas prices rise further, yet only Ford’s stock price has reacted, falling by 13.7% since the war began. Stellantis has only dipped by 1.4%, while GM has only declined by 2.4%. This seems like the market is mispricing both stocks, which should see further downside.

• Japanese carmakers could see relative outperformance: While a stagflation scenario with higher gas prices will spare no carmaker when it comes to a drop in auto demand, the Japanese Big 3 should see less of a decline, given their top-selling models like the Toyota RAV4, the Honda CR-V, and the Nissan Rogue all get around 32 mpg (8.5 km/liter). When US gasoline prices hit a high of $6.37/gallon ($1.68/liter) on an inflation-adjusted basis in 2008, the compact car segment in the US saw its market share nearly double, from 12% in the mid-1990s to 20% in 2008, while the light truck segment’s share dropped from 52% to 45%.

• Why oil over $100 now isn’t like the 2011 to 2014 period: From February 2011 to August 2014, Brent crude remained over $100, and the inflation-adjusted gas price was $5.09/gallon ($1.80/liter), yet the Detroit 3 saw surging sales of their top pickup models. Between 2011 and 2014, housing starts nearly doubled after the lows of the Great Recession, car loans averaged 4.65%, with mortgages at 3.75%, home prices were rising, and the stock market was surging. That was like Club Med compared to the current situation: housing starts are falling, car loans are at 7.52%, mortgages at 6.37%, home prices have peaked, and the S&P 500 is at its all-time high. The only thing that’s better now is the price of gas at $4.11/gallon ($1.09/liter), which is 19.3% lower than the inflation-adjusted average of $5.09 between 2011 and 2014. Given how stretched US consumers are, if gas prices rise above that rate, it could reduce US vehicle demand.

• EVs are still too expensive amid rising US affordability issues: While higher gasoline prices should be the perfect catalyst for stronger EV demand in the US, the cancellation of the $7,500 EV tax credit last September has made the US the most expensive place to buy an EV. During Q1, EV sales in the US plunged by 28% YoY, with EV market share dropping to 4.9%, from 6.3% in Q1 2025. Most of this drop was Tesla’s estimated 13.6% YoY decline in US sales, but the VW Group’s 20% drop and Ford’s 70% plunge also contributed. Among those carmakers who saw their Q1 EV sales in the US grow, Hyundai is a prime example of how hard it is to sell EVs in America without government subsidies: Hyundai’s Q1 EV sales rose by 6% YoY, but it required $6,000 to $13,000 discounts (15% to 20% off of their MSRPs) on its Ioniq 5 and Ioniq 9 electric SUVs. It’s no surprise that EVs only made up 6% of Hyundai’s Q1 sales in the US, despite its new Georgia Metaplant, which has 300,000 units of annual EV capacity.

• GM, Stellantis, and Ford look overvalued in light of the risks: The Detroit 3 saw 98% of their Q1 sales in the US made up by light truck models, and all 3 expect to see earnings growth in 2026, even on an adjusted basis. While Stellantis only generates 40% of its revenues in North America, it made up 225% of its 2025 operating losses. GM saw 90% of its 2025 Auto gross profit from its North American operations, while Ford is estimated to be closer to 100%. Unless there’s a miraculous recovery in US housing starts (which influence full-size pickup sales), or gas prices drop to pre-war levels, the downside to 2026 earnings guidance and consensus estimates is over 60% for the group. This would put GM’s consensus 2026 P/E multiple of 6x to 24x, Ford would go from its consensus of 4x to 28x, while Stellantis would go from 2x to losses (more details on the major carmakers’ 2026 outlook below).

• Toyota and BMW are the best long hedges in an auto long/short book: It’s hard to buy cyclicals in a recession, especially if it’s the first in 17 years or, even worse, the first stagflation phase since the oil shock of 1979. That being said, using “long hedges” like Toyota and BMW against shorts in Tesla and the Detroit 3 is my preferred way of hedging my automotive portfolio (see details on each stock in the last section of this report, and refer to Figure 7 in that section).

• Hyundai is the “Tesla of Asia”, but without as many lies: Hyundai’s stock was up 125% YTD before the war in Iran began, but is still the best-performing auto stock in the world after its sell-off, with 71% YTD gains. Their auto division missed Q4 2025 estimates by 38%, but the stock nearly doubled after Hyundai announced that its Boston Dynamics unit will deploy its “Atlas” humanoid bot to assembly lines at its new Georgia “Metaplant” from 2028. If you enjoy wild speculation that’s somewhat “based”, read the section below about Hyundai’s plans for “Robots as a Service (RaaS)”, and a potential $100 billion IPO of Boston Dynamics. This could be the perfect hedge against Tesla, which hit a wall in 2025 after its Optimus humanoid engineers quit, followed by the same exit of talent at xAI, which was supposed to be the “brains” of Optimus.

• Tesla is worth $26 per share: Details on why Tesla should implode this year at the end of this report, including my sum-of-the-parts estimates. Q1 2026 earnings next week will miss wildly due not only to huge incentives on Tesla’s EV sales, but more so because of Tesla Energy seeing a huge increase in its LFP cell costs, coupled with a spike in US tariffs on Chinese imports of LFP cells from 10.9% to 28.4% form January 1st.

====================================================================

Oil Over $100/Barrel Could Damage US Auto Demand

The executives at major carmakers with large US exposure have been battered by various policies launched by the Trump Administration since January 2025:

• Tariffs of 10% to 20% were placed on imported cars and parts from April last year, causing a massive supply chain headache, as most carmakers source both cars and parts from various overseas factories. This cost the carmakers in Figure X below to see a combined hit to earnings of over $30 billion in 2025.

• All support for EV adoption was removed: After a 3-year push to roll out EVs, most of the major carmakers spent billions on new EV platforms and battery supply chains. The Trump Administration removed both the “carrot” and the “stick” for US EV adoption: the “carrot” was the $7,500 EV tax credit system (canceled from Q4 2025), and the “stick” was the CAFE and GHG laws that required all carmakers to gradually shift most of their US fleet to zero-emission vehicles (also canceled). This led to plummeting EV demand and related write-downs of $53 billion among the Detroit 3 and Honda, which made up 48% of 2025 US light vehicle sales.

• War in Iran spikes the oil price: Only a few weeks after these 4 carmakers announced their huge EV charges booked, the US attacked Iran, causing US gas prices to spike by 38% at a time when affordability problems caused a 5.8% drop in Q1 2026 light vehicle sales.

Since the war in Iran started 6 weeks ago, the 38% spike in US gas prices is the steepest among developed countries (see Figure 1), largely because gas taxes are much lower than in other countries, which can use their tax rates and duties to buffer the spike (as Germany plans to do).

This is bad news for carmakers with high exposure to the US auto market, especially those with a low-mileage fleet, like GM, Ford, and Stellantis. While the market seems to be brushing off the prospect of higher gas prices, with many commentators pointing to the period between 2011 and 2014, when Brent crude stayed above $100 for 43 months, while the US stock market surged by 53%. But if oil does indeed stay above $100/barrel and gas prices in the US rise above $5.00/gallon, what we may be witnessing now could be similar to the disruption that the high oil price caused in 2008.

Figure 1: Gasoline Prices Now vs Pre-Iran War on February 27th

Source: GlobalPetrolPrices.com.

During the Great Financial Crisis in 2008, the sudden spike in oil to over $100/barrel became the final factor behind the bankruptcies of GM and Chrysler (Ford was saved by securing $23.5 billion in financing in December 2006 via bonds and a 5-year credit facility).

The US Big 3 (now, the “Detroit 3”, as Chrysler is a unit of Stellantis) mainly had gas-guzzling pick-up trucks and large SUVs in their model portfolios, as the stalling housing market and affordability issues led to higher demand for fuel-efficient Japanese cars.

Brent crude began rising from from $55/barrel in January 2007, and hit $92/barrel by December 2007, causing the top-selling pickup truck models at the Big 3 (the Ford F-150, GM’s Chevy Silverado, and Chrysler’s Dodge Ram), to fall by 8% YoY, despite huge incentives (see Figure 2).

Figure 2: Big 3 Top Pickup Model Sales vs Brent Crude Price

Source: Company data. Top-selling models: Ford F-150, GM Silverado, and Dodge Ram.

The average gas price in the US is now around $4.15/gallon, which is lower than the 3 periods when Brent crude rose above $100/barrel on an inflation-adjusted level (see Figure 3), but there are some key differences (the numbers highlighted in blue in Figure 3 are most significant):

• Car loan rates are higher than in 2008: Unlike 2008 or the 2011 to 2014 period in Figure 3, interest rates are higher now, making annual loan payments for a Ford F-150, 17.2% of the average income in the US, or nearly as high as it was in 2008, at 17.5%.

• Higher fuel efficiency is a buffer for now: As can be seen in Figure 3, while annual loan payments on a Ford F-150 are now near highs seen in 2008, the cost of fuel is only 4.2% of the average US income, versus 9.1% in 2008. The Ford F-150 XLT 4WD only drove 14 miles per gallon (3.7 km/liter) in 2008, but has improved by 36% to 19 mpg (5 km/l) currently. While this is a huge buffer for now, if gas prices go above $6.00/gallon, the cost of driving a full-size pickup like the F-150 would likely become unaffordable to most US households.

• Why things were different in other high oil phases: Many ask why Brent crude over $100/barrel between 2011 and 2014 didn’t kill the economy, and the main reasons include the Fed’s zero-interest rate policy, employment growth, a near doubling of housing starts, rising home values, and a surging stock market. The set-up for Brent crude above $100 now is much more dangerous than in 2008: interest rates are higher now at 4.3% versus 3.7% in 2008; unemployment is lower at 4.3% versus 5.0% in January 2008; and the US stock market is amazingly at new all-time highs.

Figure 3: Cost of Driving a Ford F-150 When Oil is Over $100

Source: FRED, SSA, & Ford. Note: * Monthly loan terms adjusted to 2008’s 60 months for all periods.

The US Car Market: Filled Up and Priced Out

US Car Sales Were Already Weak Before the War in Iran: The US auto market already seems to be heading into a recession, and higher gas prices could be the final tipping point for 2 main reasons:

New car prices remain near record-high levels of $50,000, and the average car loan has an annualized rate of 9%, which is a 35-year high. Monthly loan payments of $800 are also at 35-year highs, despite the average term now being near 70 months versus 60 months in 2008, which is another sign of weakness (or desperation).

New car sales drop for 6 months in a row: It’s not surprising that demand seems to be falling off a cliff now: ever since Q3 2025, monthly car sales have dropped for 6 consecutive months. The monthly SAAR (seasonally adjusted annualized rate) has averaged out to 15.6 million vehicles, or 4% lower than total sales of 16.3 million units in 2025. Q1 2026 light vehicle sales were down by 5.8% YoY, with dealer inventory up by 17.4% YoY to 92 days of supply versus 56 days at the end of Q1 2025.

While March last year had a rush in demand ahead of Trump’s tariffs from April (which led to a 14.1% jump in March 2025 sales and an 8.4% YoY pull-back in last month’s light vehicle sales), the hurdles for January and February were low, yet sales dropped by 4.2% and 1.5%, respectively.

If the oil price remains elevated for the rest of this year, it will not only hit car demand, but it will also lead to higher input costs for carmakers, which are already suffering from higher costs related to US tariffs.

In such a scenario, carmakers with low-mileage fleets in the US should suffer, while those with higher fuel efficiency should gain market share, as was the case in 2008. It should be noted that all carmakers generated huge losses in 2008, including Toyota and Honda, but their sales fell much less than those of the US Big 3.

The Detroit 3 Should See the Most Downside

Given the current conditions and the possibilities of the oil price remaining elevated above $100/barrel throughout 2026, the carmakers with the biggest earnings downside are as follows, in order of magnitude:

1. GM

2. Ford

3. Stellantis

Those that should relatively outperform are Toyota and BMW (more details below).

While the German Big 3 carmakers have less exposure to the US market, they have over 40% of their sales (and even more of their profits) hinged to China, where Q1 2026 passenger car sales were down by 17.4% YoY. A wobbly situation in the US auto market simply makes their 2026 earnings outlook weaker. Note that the German carmakers are guiding for a recovery from massive declines in 2025 profits, but they still will likely miss guidance by a wide margin. The same goes for the Detroit 3.

Before going into the earnings fundamentals, it’s best to first look at the fuel efficiency of the major carmakers’ fleets in the US, and their earnings exposure to the US market. Figure 4 below shows the fuel economy of the major carmakers that derive a large amount of their revenues from the US market, sorted from the least fuel-efficient fleet to the best. Here are the key points:

Read more

The Bad: Tesla just released their Q1 2026 delivery numbers of 358,023 (+6.3% YoY; -14.4 QoQ), which missed consensus estimates of 365,645 by 2.1% and TroyTeslike’s 375,00 estimate by 4.5%.

The Ugly: Tesla Energy’s deployment numbers were even worse, at only 8.8 GWh, missing consensus estimates of 14.3 GWh by 38.5% and down by 15.4% YoY.

The Horrible: Vehicle production in Q1 grew by 12.6% YoY to 408,386 (-5.5% QoQ), which adds 50,363 units of extra inventory to Q4 2025’s ending balance of 83,645, pushing up Q1 inventory to 134,008, which is the highest level since Q1 2024.

The trends of all three of the above are outlined in Figure 1.

Figure 1: Tesla’s Quarterly Production, Deliveries and Inventories

Source: Tesla.

Key implications of these results:

• US sales are struggling: As outlined in my short preview of Q1 deliveries yesterday (here), sales in Europe and APAC were soaring due to heavy use of 0% interest loan programs and price cuts outside the North America (see Figure 2), where Tesla doesn’t report its monthly sales numbers. The reporting countries ex-North America represent 73% of Tesla’s quarterly deliveries on average, so I estimate that Q1 North American sales (which are 95% US deliveries) came to 93,801 vehicles, down 25.7% YoY despite the low hurdle from Model Y retooling downtime in Q1 2025, and 16.2% lower than Q4 2025. This should be bad for earnings, as Tesla’s US operations have become increasingly less profitable (Auto gross margins barely rose despite 28% sequential delivery growth in North America in Q3 2025, ahead of the expiry of the $7,500 EV subsidy, detailed in the “Notable Points About Q3 Results” section of this report).

• New EV subsidies in key markets were a huge boost to volumes: In Figure 2 below, countries in the table that are colored in blue indicate those where new EV incentives were introduced from January 1st, and, combined with Tesla’s price cuts and/or 0% interest loan programs, led to a rare increase in sales over Q4 2025. Normally, Q1 is seasonally the slowest quarter, while Q4 is the highest. I also highlighted the % QoQ numbers in Figure 2 to show how most of the countries with new EV subsidy programs were the ones to see sharp QoQ rises in Q1 deliveries. Figure 3 shows all the major markets where Tesla implemented low-interest loan programs, including the impact on sales (Tesla books the cost of its low-interest loans as an up-front decrease of revenues). Figure 3 also shows how much new EV subsidies are for those countries that introduced them from January.

• Burgeoning inventory should weigh on cash flows: Given that the spread between production and deliveries of 50,363 vehicles, this pushed up Q1 2026 vehicle inventory to 134,008, up 60% versus Q4 2025 and the highest level since Q1 2024, when Tesla saw negative free cash flow of $2.5 billion. This points to a higher possibility of cash burn in Q1 2026, as Tesla saw 8% higher deliveries in Q1 2024, average unit prices were 13% higher, and capex should be much higher in Q1 2026, as Tesla guided for at least a 135% YoY rise this year.

• Tesla Energy could see a huge decline in profits: Despite having double the production capacity this year (80 GWh) due to Tesla’s Shanghai Megapack factory being fully ramped, energy deployment of only 8.8 GWh in Q1 was down by 15.4% YoY, which is not a good sign: on the Q4 2025 earnings call in January, Tesla’s CFO warned about “margin compression” this year due to “increased low-cost competition” (this is undoubtedly from China’s Sungrow, CATL, and BYD in markets outside of the US), as well as US tariffs. LFP battery cells are the key component for Tesla’s Megapacks, and they’re all imported from CATL in China. What few analysts have reported about their 2026 outlook is that US tariffs on LFP imports from China just rose to 28.4% as of January 1, 2026, which points to a huge drag on profits in Q1 from this highly profitable division (15% of 2025 consolidated gross profits).

Read more

The Q1 2026 delivery numbers coming out of reporting countries overnight show not only year-on-year growth, but even quarter-on-quarter growth in some regions of Europe.

This could lead to Tesla’s Q1 deliveries (scheduled to be reported on April 2nd) to beat the recently published Tesla-compiled consensus of 365,645 (+2.8% YoY and -12.6% QoQ), and possibly even TroyTeslike’s 375,000 estimate (+5.4% YoY and -10.3% QoQ) released yesterday.

While this could lead to a short-term rally in Tesla shares, given how bombed out it is (-22% from its mid-December all-time high of $490), a look at how much Tesla cut prices in Q1 via aggressive financing deals and outright price cuts shows how desperate they are to move the metal.

• New incentives in Q1 2026 are huge: Tesla launched aggressive low-interest-rate loan programs in 13 key markets (see Figure 1), which should cut average selling prices (ASPs) by $4,827 per vehicle. All but 2 of these markets saw Tesla launch 0% interest-rate programs, while the other 2 were at 0.99% (US and Italy).

• Strategic price cuts to qualify for EV incentives: Tesla cut prices by around $4,000 per Model Y in 4 countries in order to qualify for EV subsidies based on a maximum price cap, which Tesla exceeded without the price cuts (see Figure 1).

• Further sales of the stripped-down, affordable “Standard” trims: The “affordable” versions of the Model 3 and Y weren’t available in most countries until Q4 2025, but now are flooding the markets globally (except for China, for some reason). Given that these stripped-down variants are 14% cheaper than their unstripped versions, sales mix in Q1 2025 should deteriorate both sequentially and especially on a year-on-year basis.

• EV subsidy programs hiked Q1 deliveries massively: In Thailand, where EV subsidies were set to get cut from February, a rush in demand led to Tesla registering (but not selling) 5,800 cars, which is more than the 5,049 deliveries Tesla saw in 2025. In France, Germany, and Spain, new EV incentives worth $4,722 helped spike Tesla’s sales in those markets.

• Q1 2026 earnings should be bad: With an average price cut of around $4,000 in Tesla’s major markets, earnings could be much worse than expected (consensus EPS is $0.40 versus my estimate of $0.10).

• The ultimate question that no one has answered: Why were Tesla’s COGS/unit in Q4 2025 down by 4.2% versus Q3, despite the 16% sequential drop in deliveries? Please chime in if you have a good guess, but costs must rise eventually as lithium prices are now 2x higher than last year.

Read more

This is a short report on the imminent plunge in Tesla’s “free-money” regulatory credit sales. The Environmental Protection Agency (EPA) nullified greenhouse gas (GHG) credit sales in the US last month, which should fully impact Tesla’s earnings from Q2 (good report on why the EPA did this here). ZEV credit sales in Europe should also plunge from Q1 2026, after Toyota and Stellantis left Tesla’s “credit pool” from the start of this year.

While Tesla is down by nearly 30% since hitting its all-time high of $490 three months ago, Figures 4 through 9 show how much deeper the downside could be.

• 28% of Tesla’s 2025 earnings decline came from the drop in regulatory credit sales, from $2.76 billion in 2024 to $1.99 billion last year. In light of Trump’s EPA having ended the GHG credit sales scheme last month, there should be another $1 billion headwind for 2026 earnings. This increases the chances of high cash burn, as regulatory credits made up a record 84% of Tesla’s 2025 EBIT and 32% of free cash flow.

• GM warned that GHGs were 78% of its credits and face impairment: US regulatory credits for Tesla are thought to have made up 70% to 80% of total credit sales before the ZEV credit system ended in Q3 2025, but it’s impossible to tell due to Tesla’s lack of disclosure. However, it has become clear that GHG credit sales in the US were much higher than ZEV credit sales at Tesla: GM warned in its 2025 10-K that, if the EPA canceled the GHG credit scheme, it would face an impairment of $1.1 billion, or 78% of its total credit pool of $1.4 billion. Below, a look at Tesla’s record $2.76 billion of credit sales in 2024 shows why GHG credits were likely more profitable at Tesla as well.

• Toyota and Stellantis left Tesla’s credit pool in Europe: On February 27, the EU disclosed that Toyota and Stellantis were no longer members of Tesla’s 2026 credit pool. While it’s hard to say how much this reduces Tesla’s regulatory credit sales, Chrysler (now part of Stellantis) initially signed a $2 billion contract in 2019 to be in Tesla’s credit pool. Stellantis left Tesla to join the credit pool of its partner, Leap Motors, a rapidly growing Chinese EV maker, in which Stellantis owns a 51% stake of a joint venture for Leap Motors’ international sales (mostly in Europe). The departure of Toyota, Stellantis, and Leap Motor reduces Tesla’s credit pool by 80% in volume terms.

• Tesla’s regulatory credit obligations already sank by 82% in 2025: In its 10-K that came out in January, Tesla stated that Trump’s budget bill last year led to an 82% YoY decline in long-term credit sales obligations (those with contracts of over 1 year), from $4.68 billion at the end of 2024 to $0.84 billion at the end of 2025. The exit of Stellantis and Toyota could be part of this decline, as Tesla would’ve known by late 2025 that both giants were leaving its EU credit pool. It could also be why long-term obligations dropped from $3.27 billion at the end of Q3 2025 to $0.84 billion in Q4 2025. But this surely didn’t factor in the end of GHG credit sales, which only started from this month.

• Broken chart with no valuation support: While Tesla has dropped by nearly 30% since its all-time high of $490 on December 16th, the chart looks bearish. And, at 170x consensus EPS estimates of $2.08 in 2026, and a more sober 221x trailing 12-month EPS, there is zero valuation support. Figures 4 through 7 show how much further Tesla could fall from a chart perspective, while Figures 8 and 9 show how overvalued Tesla is relative to the world’s top 15 carmakers, as well as its “colleagues” in the Magnificent 7.

====================================================================

Tesla was already set to see weak earnings this year, despite the rosy outlook on Wall Street, which sees 8% revenue growth and a 25% rebound in EPS to $2.02 this year (I’m forecasting an 11.6% drop in revenues and EPS of -$0.30). The reality is that, without any new model launches and little hope of Tesla’s robotaxi business taking off (see why in this report about mounting FSD troubles), Tesla must cut prices further to keep its factories running. And this is exactly the case so far in Q1 2026, with more aggressive low-interest rate loans and further price cuts in key markets. But things could get worse.

With increasing chances that the war in Iran will cause an economic slowdown, Tesla could see even weaker demand than expected, which would be catastrophic in light of 2026 being the first year with no more ZEV or GHG credit sales in the US (roughly 70% of global regulatory credit sales of $2.76 billion in 2024), while EU credit sales should also decline.

The ZEV credit sales system in the US ended in Q3 2025 after being neutralized by Trump’s July 2025 tax and spending bill, while his EPA chief eliminated the GHG credit scheme in February. If Tesla sticks to its guidance of 2026 capex being “in excess of $20 billion” (or 2.5x higher than 2025 capex and 42% more than 2025 operating cash flow), negative free cash flow (FCF) becomes a sure thing. My estimates are already bearish at -$15 billion, or 23% of Tesla’s cash and equivalents at the end of 2025.

Figure 1 below is a good visual image of just how bad Tesla’s core earnings have been since their peak in 2022, with 85% of last year’s operating profit made up by regulatory credits, which exceeded the 79% level in 2020, Tesla’s first profitable year. For reference, Figure 2 shows the breakdown of these numbers since 2020.

Figure 1: Tesla’s Core Profits Hit a 5-Year Low

Source: Tesla.

A few key points about the numbers used in Figure 1 and listed in detail in Figure 2 below:

• ZEV and GHG credits have 100% profit margins for Tesla, most of which came from the US (Tesla doesn’t disclose the breakdown between ZEV or GHG, as well as regional sales, but roughly 70% are believed to have come from the US). While Tesla still earns ZEV credits in China and in Europe, the amount in China is small, as EV subsidies have been reduced, while Tesla’s key customers for credits in Europe left its credit pool at the end of last year. The impact of lower credit sales in Europe will hit in Q1, but the hit from zero GHG sales from Q2 should be much harder, as explained below.

• GHG credits in the US were likely quite big, and could be the main reason why Tesla’s 2024 credit sales of $2.7 billion grew by 54% YoY (see details in Figure 2). It’s impossible to know how much Tesla’s GHG credit sales in the US came to in 2024, but the huge $973 increase from 2023 levels could be linked to GHG credits: Tesla generated a record 34 million metric tons of GHG credits from sales of its 2023 model year in the US (details here), which was a 78% YoY increase from 19.1 million metric tons of GHG credits from its 2022 model year (model years in the US usually start from Q3 of the previous year, so the 2023 model year with record GHG credits came from US deliveries between Q3 2022 and Q2 2023).

• GHG credits gone from Q2 should hit profits hard: The 34 million metric tons of GHG credits that Tesla earned via its sales of its BEVs in the 2023 model year came at the perfect time, as the industry booked a deficit of 42 million metric tons, which likely hiked the price of GHG credits in 2024. This likely made up most of the $973 million YoY increase in Tesla’s record regulatory credit sales of $2.76 billion in 2024. Another sign that GHG credits were a much larger chunk of Tesla’s total credit sales is GM’s warning in its 2025 10-K: GM stated that, if the EPA removes GHG regulations (which it did in February), $1.1 billion of its total $1.4 billion of credits acquired would face impairment. The fact that 78% of GM’s 2025-end regulatory credit pool was GHG credits is telling: GM sells most of its cars in the US, which implies that Tesla likely saw more profits from GHG credit sales in the US than its ZEV credit sales, which is why profits from Q2 onward could suffer.

• IRA manufacturing credits remain: In Figure 2, one can see that IRA manufacturing credits have grown dramatically since the Biden Administration created the standard to speed up EV adoption in the US. While Trump got rid of Biden’s $7,500 EV subsidies after Q3 2025 (which led to pulled-forward demand that increased US deliveries by 28% sequentially in Q3 2025), the IRA manufacturing credits remained, most likely on Musk’s request. These credits are awarded to makers of battery cells and battery packs in the US, which lowers production costs, so they don’t show up in revenues like regulatory credit sales. As can be seen in Figure 2, IRA credits peaked for Tesla’s auto division in 2024, given the decline in US output, but reached new record highs for its Energy business, due to huge demand for its Megapack energy storage systems. On a combined basis, IRA manufacturing credits added $474 million to 2023 profits (21% of total credits) and jumped to $1.69 billion in 2025, making up 46% of total regulatory credits.

• Energy division’s Megapack sales drive credits from here: While the auto portion should drop in line with lower EV output in the US, Tesla’s Megapack sales have room to rise by 124%, given the new Texas factory, which doubles Tesla’s US annual capacity to 80 GWh. However, the IRA credit scheme is scheduled to be phased out from 2030, in 25% increments per year until they hit zero in 2032. Given that much of the strong demand seen in the US for Tesla’s Megapacks comes from AI data centers (nearly all the top players use Tesla Megapacks), any slowdown before 2030 could hurt. Last year’s $1.12 billion of IRA credits made up a whopping 30% of gross profits in Tesla’s Energy Division, which would’ve only seen a 19.6% gross margin without these credits, rather than the 29.8% it reported.

Figure 2: Breakdown of Tesla’s Profits & Regulatory Credits

Source: Tesla.

While there’s lots of talk about how overvalued Tesa’s stock is, especially on a trailing 12-month basis (many are starting to doubt the Street’s rosy forward outlook), stripping out total regulatory credit contribution to EPS is likely more valid: while US ZEV credit sales were gone as of Q4 2025, this year’s Q2 should see zero sales of the highly profitable GHG credits, on top of a huge drop in EU credits, in light of Toyota and Stellantis having left Tesla’s credit pool at the end of 2025.

• Consensus estimates of $2.08 in 2026 EPS (+27% YoY) gives Tesla a 185.5x forward P/E multiple.

• Based on Tesla’s 2025 EPS of $1.60, its trailing 12-month P/E comes to 225.3x, which is often quoted these days.

• But when regulatory credits of $1.99 billion are removed, Tesla’s 2025 adjusted EPS drops to $0.50, causing the trailing 12-month P/E multiple to jump up to 772.3x.

Figure 3 shows Tesla’s non-GAAP net profits (ex-credits) since 2021, with the trailing P/E multiple calculated by the average share price in the following year. As can be seen, Tesla’s earnings have fallen sharply since peaking out in 2022, but the market continues to award Tesla an “AI premium”, despite zero signs of Tesla having any edge in AI:

• No robotaxis for a while: Tesla’s cars still can’t drive themselves without human supervision, and their Full Self-Driving (FSD) technology is under a deep probe by US regulators who want to know why FSD’s “AI” causes crashes by willfully choosing to run red lights, drive in opposite lanes, etc.

• No humanoid bots for a long time as well: Given that Musk’s AI start-up, xAI, was supposed to be the brain of Tesla’s Optimus bot, Tesla’s “AI” premium makes even less sense, as xAI lost all of its co-founders, aside from Musk. As an AI venture, xAI is literally forced to go back to the drawing board. The Optimus project is very similar: its team leader and most key staff quit last year, after problems with development.

In fact, Tesla should lose most of its AI premium at this pace. Not only is Waymo scaling much faster than expected (500,000 driverless rides per week in 10 cities, with a total of 15 cities in 2027), but the number of new entrants is rapidly rising, with Amazon’s Zoox (5 cities in 2027) and Uber’s various partnerships with VW, Nvidia, Lucid, Rivian, and Nuro, which all launch in 2027. With humanoid bots, it’s even worse, given that Tesla was already late to the market, and existing players like Boston Dynamics and Figure AI are poised to dominate the US market, while 100s of Chinese rivals race for supremacy at the same time. Tesla is being left behind, which begs the question of why its stock deserves an “AI premium” at all.

Figure 3: Tesla’s Non-GAAP Net Profit vs its TTM P/E Multiple

Source: Tesla, Bloomberg, & MH research.

As is clear in Figure 4, Tesla’s stock chart is looking bad after dropping 27% from its all-time high of $490 on December 16th. This is largely due to Tesla still having safety drivers in its “robotaxi” fleets in Austin, despite Musk having promised at least 5 times last year that the drivers would be removed by year-end. Two near-term catalysts could cause further downside:

• Cybercab launch in April: This should disappoint, as the highly anticipated launch of the “Cybercab” in April will be met without a functional FSD, which means it can’t be driven, since Musk has insisted that it won’t have a steering wheel or pedals. Using safety drivers for the Cybercab would be awkward, given that it only has 2 seats. Another bad sign: 3 leaders from the Cybercab program left Tesla over the past 5 weeks, despite the April launch schedule.

• SpaceX IPO in June: The SpaceX IPO, which is scheduled for June, should be another negative catalyst, as Musk is aiming to raise a record $75 billion with a huge 30% allocation for retail investors (normally, retail only sees 5%-10% of an IPO). At 39% of shares outstanding, Tesla has the highest retail investor base among mega-cap stocks. If even a fourth of these investors swap their Tesla for SpaceX, it could tank Tesla without the usual dip-buyers, as it would amount to 5.4 days of average trading volume (see my detailed report on the SpaceX IPO here).

Read more

These are the main topics covered in this report:

• SpaceX is more sexy than Tesla: The SpaceX IPO could tempt retail investors to sell their Tesla shares for SpaceX, which is being hyped to see much higher earnings growth than Tesla, with more exciting themes like AI data bases in space.

• Retail selling could crush Tesla’s stock: At 39% of shares outstand…

Read more

On Saturday, Elon Musk announced that Tesla, SpaceX, and xAI will jointly invest $25 billion in a “Terafab,” a semiconductor fab designed to match the output of the world’s largest chip plants by 2030 and deliver 50 times the current global AI compute power thereafter.

This comes just weeks after Musk justified SpaceX’s $250 billion buyout of xAI, his cash-incinerating AI start-up that lost 10 of its 12 co-founders in recent months.

The question is: why is Musk announcing yet another insane project in such a short time span? Obviously, because he thought it could boost the value of SpaceX, which is planning a $50 billion IPO at a $1.75 trillion valuation (Polymarket now has the odds of SpaceX’s first-day closing market cap of $2 trillion at 57%, up from 45% pre-announcement).

But it could also provide the perfect excuse for Tesla to raise cash amid a tough outlook for 2026, during which Tesla’s CFO said Tesla would see capex “in excess of $20 billion”, which is 43% above last year’s operating cash flow of $14.7 billion.

The CFO listed the following items as the need for over $20 billion in 2026 capex:

1. New lithium refinery

2. New LFP factory using old CATL equipment

3. New Megapack energy storage factory

4. Cybercab start of production

5. Semi truck start of production

6. Converting the Model S/X production line into Optimus lines

7. AI compute infrastructure

8. Maintenance capex for existing factories

He emphasized that all of the above investment needs excluded plans for the Terafab and solar cell manufacturing.

It should be noted that items 1 through 5 are all mostly paid for, as they’ve already started operating or will do so soon, so it’s hard to use them as an excuse for doubling capex, plus alpha. This is why Terafab is the perfect new carrot to dangle in front of investors while Tesla goes through its “transitional phase” from a carmaker to who-knows-what.

Here are my main doubts about Musk’s Terafab project:

• Putting more carts in front of the horse: Just like Tesla didn’t need to develop its Cybercab without FSD being fully autonomous, there’s no need for Terafab-levels of chip supply if neither Tesla robotaxis nor Optimus are even close to being rolled out (see my recent report on how badly FSD and robotaxis are functioning here). The same goes for Musk’s ambitions for orbital AI databases: SpaceX can barely get to the moon, while xAI’s “intelligence” as an AI is subpar compared to rivals.

• SpaceX may not be as profitable as its pre-IPO pitch implies: Given Musk’s claim that 80% of the Terafab’s output will be used by SpaceX and xAI in space (the other 20% is for Tesla’s robotaxis and Optimus), it’s strange that SpaceX can’t cover the entire project. Musk said that Terafab’s first phase will cost around $25 billion, which one would think SpaceX could handle after its expected $50 billion IPO, but maybe SpaceX has larger cash needs.

• Musk fails badly when straying from auto manufacturing: Musk has a horrible track record in his non-automotive ventures (see a list of Musk failures in Figure 1). He thought Tesla could take on battery cell manufacturing, but his 4680 cell concept was a failure (a Korean supplier wrote down its order contract from Tesla by 99% recently), and his Dojo project saw its original team disband last year. Given how semiconductor production is much more complex (a 2nm fab requires roughly 3,000 to 5,000 specialized PhD-level engineers for lithography, plasma etching, chemical deposition, etc.), it’s doubtful that this Terafab venture will succeed, which is why it could be more of a stock pump for Tesla and pre-IPO SpaceX.

Figure 1: Musk’s Biggest Failures So Far

Source: Tesla & MH research.

Skeptical Points About Musk’s Terafab Plans

Below are the key Terafab specs and goals outlined on Saturday, and why most of them are “pie-in-the-sky” ideas typical of Musk:

Read more

I was about to publish a quick update on how bad the monthly crash data on Tesla’s FSD and Austin “robotaxis” were, when news came out that NHTSA had raised one of its FSD probes from “PA” to “EA”, which is serious.

Tesla’s Preliminary Analysis, being raised by NHTSA to an Engineering Analysis, is like a Grand Jury’s ruling that there’s “probable cause” to move forward to trial. My March 6th report (here) is a good primer on why FSD was regressing in its capabilities.

The update below was written just before NHTSA announced that it was “strengthening” its probe on FSD crashes related to bad vision. It was meant to be (yet) another reminder that FSD and Tesla’s robotaxi fleet would definitely not launch this year (as predicted in this report).

A few points on today’s NHTSA announcement, and then below that, the key points from the report I had written on Tesla’s bad FSD report from NHTSA on Monday:

1. Killing FSD makes Tesla cheaper if SpaceX wants to buy it: Details in my next report regarding the SpaceX IPO, but I think that Musk could try to make SpaceX his next “piggy bank”, as a Tesla crash over “no robataxis again this year”, could leave him high and dry. While there was huge speculation on SpaceX merging with xAI or Tesla earlier this year, SpaceX ultimately absorbed xAI on February 4th, at a highly unreasonable valuation of $250 billion for xAI. Two weeks later, roughly 6 of xAI’s engineers left, including the last 4 of its 12 co-founders. Given that xAI has negative free cash flow of $13 billion versus an estimated $2.5 billion of positive FCF at SpaceX (excluding acquisitions), potential investors may be asking for a bigger discount on SpaceX shares. This could lead to a lower Tesla share price, if those who were betting on being bought out by SpaceX at current valuations sell due to worries about no robotaxis, or those who were only in Tesla for exposure to Elon Musk sell earlier ahead of the SpaceX IPO. It should also be noted that consensus SpaceX estimates for 2030 revenues have plunged by 33% to $60 billion in the past week, while 2030 EBITDA forecasts dropped by 28% to $35 billion (which still leaves SpaceX at a lofty 2030 EV/sales of 29x and an EV/EBITDA of 50x).

2. Even without a buyout by SpaceX, Tesla shares should dive lower: Much of the reason that Musk chose June for his SpaceX IPO is to suck out as much capital from the markets as possible before OpenAI and Anthropic go public in the 2H of 2026. If SpaceX actually does raise $50 billion in cash by July (which is an amount equal to 73.5% of total money raised from IPOs and SPACs last year), lots of Tesla investors who want exposure to SpaceX, need to sell Tesla to swap into SpaceX. At the same time, lots of ETFs that hold Tesla as a “space stock”, will also need to sell their Tesla shares in order to be in compliance with whatever their single-stock exposure mandates are. Hence, the SpaceX IPO posing a selling event for Tesla shares, which most holders have seen flat returns on over the past 4-5 years.

3. Conclusion is that this NHTSA FSD probe being elevated to an “Engineering Analysis” level is seriously bad news. As explained in my March 6th deep dive on the flaws of FSD (here), roughly 93% of Tesla’s current $1.3 trillion valuation is linked to its “robotaxis” overtaking Uber and Lyft in the global ride-hailing market. The fact that Tesla is still having problems with its much criticized “vision-only” ADS, doesn’t bode well for those expecting Tesla to scale its “robotaxi” fleet.

===================================================================

The market is awarding Tesla’s “self-driving” technology a $1.17 trillion valuation, or $350/share, while the stock is theoretically worth only $12/share on a 2026 EV/Sales multiple of 0.45x (sector average). My last report from March 6th is a key guidebook on how to see why Tesla’s FSD is not nearly worth what the market values it at, and likely more of a liability (see details here).

The crash data released on Monday for Tesla’s Level 2 ADAS incidents, as well as its “robotaxi” mishaps, point to FSD, under human supervision, is still 3x more likely to have an injury-related crash than human drivers (see Figure 1). This is not the “disruption” that Tesla fans expected FSD to unleash on the global ride-hailing market.

And today’s announcement from NHTSA that they’re “upping” the intensity of their FSD probe, shows that the crash data Tesla reports to NHTSA (even after Musk tried to “water down” the rules of reporting) renders it far from being as reliable as Waymo, whose robotaxis are 80% safer than human drivers. Below are the 3 key points from this week’s NHTSA crash reports:

Read more

Tesla’s share price closed below its 200-day moving average on Friday for the first time since September, and the chart now looks really broken (see Figure 3).

There’s a palpable nervousness among Tesla fans about the Cybercab, which is scheduled to start production next month but still lacks a permit to be sold as designed, without a steering wheel or gas/brake pedals. Because this is the only new product Tesla has within its aging model lineup, tensions are high.

The only thing that had held Tesla’s stock above $400—a psychologically critical level, which was breached 4 times but pumped back up since last September—is the continuous high hopes for robotaxis.

The market is still largely duped by Musk’s lies about Tesla’s ability to launch a paid driverless ride-hailing service, as the mounting death toll and serious injuries have rendered Tesla’s “Full Self-Driving" software 4x less safe than humans (see details in last week’s report).

• Deliveries outside of North America are up 11% YTD: Most countries except for North America have already reported Tesla’s registrations, and they’re up by 11% YTD, after a 35% YoY surge in February (see Figure 1).

• Q1 2026 has a low hurdle due to weeks of downtime last year: Giga Shanghai, Tesla’s largest factory, saw over 3 weeks of production downtime from January to February 2025 due to retooling for the refreshed Model Y. This makes Tesla’s February sales growth of 43% YoY in China seem spectacular, but it’s actually +28% YoY on a daily-sales basis, with revenues of $1.8 billion down by 7.7% YTD in light of 8.5% price cuts. Tesla’s European operations are even worse, despite having longer downtime in Q1 2025, with deliveries flat YTD and prices down by 8% from Q4 2025 levels.

• Korea is carrying Q1 2026 global deliveries: Tesla cut its Model 3 and Y prices by 16% on average, which, along with Korean EV subsidies, lowered the price of the cheapest Model Y to $25,354 versus $39,990 in the US. This has led to 9,834 deliveries YTD in Korea, or 16% of 2025’s total of 59,916 (see Figure 2). Tesla is selling these cars at cost or lower in order to keep its Shanghai factory from falling below 80% capacity utilization.

• The stock chart is broken and valuations remain extremely high: After having struggled for 5 times since September to stay above $400, Tesla’s closing price of $391 on Friday was the first time since then that it closed below its 200-day moving average (see Figure 3). Moreover, the usual call buying doesn’t seem to be as effective as last year, as it only worked in concert with "robotaxi” catalysts (see Figure 4). This is also the first time in a while that the put/call ratio declines with the stock price, rather than moving in the opposite direction up to now (see Figure 5).

• Tesla’s market cap is 28% higher than its top 15 rivals, combined: Tesla’s EV/sales ratio is 15.35x versus the sector average of 0.47x, while its P/E multiple is negative versus a sector average of 11.2x. Despite having one of the worst earnings outlooks in the sector, Tesla’s current market cap of $1.3 trillion is 28% higher than the $1.01 trillion market cap of its top 15 rivals, combined (see Figure 7).

• Tesla’s 12-month price action is 2nd best among the Magnificent 7: Despite having the lowest profitability among its Mag-7 rivals and negative earnings growth, Tesla’s share price over the past 12 months is still up 63%, or second only to Google’s 83% (see Figure 8). While Tesla usually outperformed rising markets and underperformed falling markets through 2024, this trend was broken in 2025, and hasn’t recovered so far this year (see Figure 6). The coming SpaceX IPO looks to raise $50 billion (74% of total IPOs + SPACs in 2025), and this should cause forced selling for investors looking for the next “10 bagger” and those who can only be exposed to one “Elon Musk” stock. Given its sky-high valuations versus real tech companies in the Magnificent 7 (see Figure 9), Tesla could plunge ahead of the SpaceX IPO if there hasn’t been an official launch of its driverless robotaxis.

====================================================================

Figure 1: Low 2025 Hurdle Drives Sales Growth Outside of the US

Source: Local car registration agencies.

Korea is Tesla’s New “Dumping Ground”

Tesla’s huge growth in South Korea during January and February 2026—with nearly 10,000 units delivered, nearly quadruple the previous year’s volume—is a result of aggressive pricing to uphold healthy operating rates at Giga Shanghai, Tesla’s most profitable factory.

Tesla strategically lowered the price of the Model Y RWD to ₩49,990,000 ($34,349), precisely undercutting the government’s ₩53 million price cap to qualify for 100% of the newly announced 2026 EV subsidies ($4,300 on average). This move, combined with a new ₩1 million ($687) “cash for clunkers” deal for people trading in their ICE vehicles, allowed Tesla cars a deal many Koreans couldn’t refuse.

Figure 2: Tesla’s Monthly Deliveries in Korea May Be Peaking

Source: KAIDA.

China Sales Drop Despite Low Hurdles & Higher Incentives

Read more

Many Tesla bulls are scratching their heads these days over why Tesla hasn’t increased the number of driverless robotaxi rides in Austin, where Musk promised to remove the safety drivers by the end of 2025.

While it’s been widely reported that the Austin robotaxi fleet has logged 14 crashes since July over only 820,000 miles traveled (4x less safe than human drivers), little is known about FSD crashes among Tesla owners.

This is because Tesla redacts this data from crash reports, as it’s unsavory and would likely put Musk’s promises of robotaxis for “half the population by the end of the year” further into question.

This report sifts through the data by using filters from what isn’t redacted and finds that FSD used by Tesla owners is becoming increasingly dangerous, not safer than human drivers, as Tesla claims in its safety reports. These are some of the key points:

• Tesla robotaxi crash data shows that humans are 4x safer: Data published by the US safety regulator in January showed that human drivers are 3.9x safer than a Tesla “robotaxis” —which have safety drivers in the front seat—while Waymo driverless cabs are 4.7x safer than Tesla’s robotaxis.

• FSD crash data is almost as bad and quickly deteriorating: While Tesla forces US regulators to redact most details in their FSD crash reports, isolating FSD-specific situations shows that Tesla owners—and innocent bystanders—are in even greater danger. Since July 2021, there’ve been 28 fatalities and 555 serious injuries related to FSD, making it 1.9x more dangerous than human drivers. Ford’s Pinto was recalled in 1978 after having 28 deaths from its misplaced gas tank, causing the car to catch fire after getting rear-ended.

• Musk continues to insist that driverless paid rides have launched: In most US states, robotaxi permits require data proving that the automated driving system (ADS) meets Level 4 autonomy standards. Tesla has yet to submit such data anywhere, yet Musk and his executives at Tesla keep promoting Tesla’s driverless rides in Austin (which are rarely available). Given its track record of being sanctioned for FSD consumer fraud in California last year and its horrible track record of crashes in Austin, it’s highly unlikely that Tesla can expand paid driverless rides any time soon.

• Cybercab’s start of production in April will be faked: Since the Cybercab has only 2 seats, and no steering wheel or gas/brake pedals, it must have FSD in order to be functional. Without FSD, Cybercab is useless, but even if FSD were fully autonomous, the Cybercab still can’t drive on US roads: its lack of a steering wheel and pedals violates US safety standards. Many investors are betting that the Cybercab not only launches in April, but that it starts delivering driverless, paid rides. This is close to impossible unless Congress changes the federal safety standards, *and* FSD makes a 20x improvement in reliability.

• Regulatory probe weighs heavily on FSD: Section 5 below has details on how much trouble Tesla is in with the US safety regulator. It began with Tesla stalling crucial FSD crash reports, and is now in a situation where Tesla is being forced to explain 8,313 extra FSD crash incidents that they didn’t know existed. If even 25% of these newly discovered 8,313 incidents become verified FSD crashes, it would further skew its safety as being less safe than human drivers. The probe is poking at the heart of FSD, not the crashes, which could be devastating for Tesla. If Tesla can’t prove that its AI isn’t purposely violating traffic laws, a recall of FSD is unavoidable.

• Current valuations can’t hold up without driverless robotaxis: On a sum-of-the-parts basis, Tesla should only be worth $27 per share, yet its current stock price is $397. The difference is worth $1.2 trillion dollars, and it’s largely attributed to high hopes of Tesla disrupting the global ride-hailing markets with its low-cost robotaxis. If there are no signs of increased paid, driverless rides by Q2 results, it’s more than likely that the oxygen in this epic bubble begins to seep. Tesla currently trades at an EV/sales ratio of 16x and a negative P/E multiple, versus the auto sector average of 0.47x and 11x, respectively.

• This is a data-heavy report with 9 tables, as well as 8 video clips of Teslas on FSD in action.

====================================================================

Introduction—The $1.2 Trillion Robotaxi Illusion

Many Wall Street analysts are forecasting Tesla’s launch of robotaxis by the end of this year and that they’ll gradually scale to high levels by the end of next year. This implies that they’re assuming FSD is capable of Level 4 autonomous driving, which public data from NHTSA fully negates.

I’ve queried one Wall Street analyst about his claims of Tesla being able to scale its robotaxi fleet, and it quickly became clear that he wasn’t analyzing the monthly crash data published by the National Highway Traffic Safety Administration (NHTSA). Moreover, he had no knowledge about the complex regulatory landscape, which makes it impossible for Tesla to operate a driverless ride-hailing service without proof of FSD being a Level 4 automated driving system (ADS).

This report is a product of weeks of sifting through reams of (mostly redacted) crash data reported by Tesla to NHTSA. Aside from showing why Tesla’s self-driving capabilities are still not safe enough for fully autonomous robotaxis, or for use by Tesla owners, this report also has 8 video clips of dangerous near misses.

Musk knows that Tesla’s Full Self-Driving (FSD) technology is a long way off from achieving his endless promises of providing driverless rides for “half the US population”. As detailed below, Tesla cars on FSD have led to 28 fatalities and 585 serious injuries. This is why Tesla has yet to scale its driverless robotaxis in Austin, which has 8 “unsupervised” variants, but only 1 is truly in service, and it hasn’t been seen for over 2 weeks.

Musk nevertheless continues to inflate the illusion of Tesla’s imminent robotaxi launch, promising—for the second straight year—driverless rides for "half the population of the US by the end of the year”.

These empty promises are the key reason why Tesla trades at a $1.4 trillion market cap, when its sum-of-the-parts valuation is only $94.3 billion, or $27 per share. Tesla’s auto business should only be worth $14 per share, based on the average 2026 EV/sales multiple of 0.47x in the auto sector, while its Energy division is only worth $13 per share based on its rivals’ 2029 average PER of 15x.

The robotaxi illusion that Musk has created is, therefore, being awarded a $1.2 trillion valuation, or $350 per share, for something that doesn’t exist. This is both preposterous and stupendous, at the same time, and it explains why Musk has gone to such great lengths to keep the dream going (e.g., 2024’s Cybercab unveiling on a Hollywood set, 2025’s video of a self-delivering Model Y to an Austin customer, and orchestrated “driverless” robotaxi rides in the past 2 months).

But it seems like few investors or blue-chip investment banks like Morgan Stanley have done the research on FSD’s crash rate, which shows how half-baked FSD is as an ADS, and how dangerous it would be to allow FSD-driven robotaxis to scale (Tesla only has one operating driverless robotaxi in Austin, likely only available for Tesla employees or influencers who signed NDAs).

This report shows the unsavory truth behind Musk’s highly orchestrated “self-driving” illusion, and is 100% backed-up by data reported by Tesla to NHTSA. This data is the result of a rigorous multi-layered audit of NHTSA’s SGO Level 2 ADAS crash reports.

For a complete breakdown of the filtering process used to isolate FSD-specific crashes, see the Appendix at the end of this report, which is divided into the following sections:

Section 1: Musk’s “Robotaxi” Dog and Pony Show

Section 2: Behind the Hype: Mounting Robotaxi & FSD Crashes

Section 3: Tesla’s Robotaxi Crashes Highlight the Dangers of FSD

Section 4: FSD Crashes Were 70% More Dangerous in 2025

Section 5: The Mounting Regulatory Squeeze on FSD

Section 6: Tesla is 3 to 4 Years Away From Going Driverless

Appendix: Isolating FSD Crashes Among Tesla’s Redactions

SECTION 1

Musk’s “Robotaxi” Dog and Pony Show

Tesla’s stock hit its all-time high of $492 last December after a couple of its employees uploaded videos of themselves in the back seat of a Tesla robotaxi, with no one driving the car (see Videos 1 and 2 below). Tesla’s Head of AI Software, Ashok Alluswamy, tweeted “And so it begins!” which helped pump the stock.

However, aside from some very well-orchestrated PR stunts with Tesla influencers in late January (see Videos 3 and 4 below), there have been few other signs of any “unsupervised” robotaxi rides for the general public in Austin. This is highly odd, given what Musk said on the Q4 2025 earnings call in late January:

Note that he said these driverless rides were “paid rides”, which is key, as most states, including Texas, prohibit paid driverless rides unless the robotaxi operator has an automated driving system (ADS) that meets Level 4 self-driving capabilities.

Because Texas is in a transition phase with its new autonomous vehicle (AV) laws, Musk is capitalizing on this grace period with dangerous PR stunts before May 28, when the Texas Department of Motor Vehicles (DMV) can issue AV permits and start enforcing the new laws (see details here). The first test of whether Tesla’s Full Self-Driving (FSD) technology is an actual Level 4 ADS will therefore be around June, when Tesla should have obtained its AV permit from the Texas DMV.

Also note that Musk said Tesla stopped using a “chase car” behind its driverless robotaxis as of late January. This is because he sensed that retail investors didn’t get that excited after the January videos of a few influencers getting driverless rides because they were followed by chase cars (see Figure 1 below), which isn’t exactly “driverless”.

Instead of reaching a new all-time high as it did in December after Tesla employees posted their driverless rides on Twitter, Tesla’s stock rose only 4% for a day, before dropping by 12% over the following 11 trading days.

Figure 1: Chase Cars With Safety Drivers Gone as of Late January

Source: Electrek.

Video 1: A Tesla Model Y in Austin Without a Driver (Dec-14)

Source: Twitter.

Video 2: Tesla’s AI Chief Rides in a Tesla Robotaxi (Dec-24)

Source: Twitter.

And despite Musk’s claim that the January driverless rides were “paid rides”—accepting money for unsupervised rides is effectively against new AV laws in Texas—one of the influencers all but admitted that he was tapped by Tesla to participate in the “paid driverless ride” event (see Figure 2).

Figure 2: Tesla Influencers Likely Didn’t Pay For Their Rides

Source: Twitter.

Video 3: Tesla Fanboy In a Driverless Robotaxi (Jan-22)

Source: Twitter.

Video 4: Tesla Influencer In a Driverless Robotaxi (Jan-22)

Source: Twitter.

Tesla’s “Robotaxi Show” Becomes More Elaborate and Costly

Musk’s “robotaxi” dog and pony show started wholeheartedly in October 2024, when Tesla unveiled the Cybercab at an event they called “We Robot” (horrible wordplay on an Asimov classic). From this one-time event at Warner Bros. studios, Tesla launched its much more expensive “robotaxi” production in Austin last June and in the Bay Area last September—all with safety drivers in front.

Currently, there are only 45 robotaxis in Austin, while the Bay Area has a full-fledged Uber-type ride-sharing fleet of 315 Model Ys. At an estimated annual cost of $150,000 per “robotaxi” (drivers’ salaries, data processing, insurance, cleaning, etc.), Musk’s dog and pony show is now at a $55 million annual run rate.

Given that this keeps the stock elevated at excessively rich valuations, this is money well spent. The question is how long the market will award a carmaker in decline such nose-bleed valuations for a “robotaxi” business that doesn’t drive itself. Especially if FSD is deteriorating.

Stark Difference Between FSD Perception and Reality

The sections below explain just how severe the recent surge in crashes has been, both among the Austin robotaxi fleet and the Tesla consumer FSD fleet in the US.

The key points, which are based on a deep dive into crash data reported by Tesla to NHTSA, are in stark contrast to what Musk is trying to portray FSD’s capabilities as in Videos 1 and 2 above:

• Austin robotaxis are statistically less safe than human drivers: In last month’s ADS crash report from NHTSA, Tesla’s Austin robotaxi fleet racked up 4 extra crashes in just 4 weeks since mid-December, 3 of which were in the first 2 weeks of January. Despite having safety monitors in front, Tesla’s robotaxis now have a track record of 1 crash per 58,571 miles (94,261 km), making human drivers 4x safer (see Figure 3 below).

• FSD crashes have become more dangerous: The estimated number of crashes by Tesla drivers using FSD only rose by 5% to 398 in 2025, while, adjusted for fleet miles traveled, the figure showed a 47% improvement from 0.17 incidents per million miles (IPMM) in 2024 to 0.09 in 2025. While the blended IPMM is 0.09, this figure is heavily diluted by low-risk highway driving and does not reflect the much higher failure rate in urban environments. What the regulators are up in arms about, however, is the ratio of fatalities: FSD had 1 death per 36 crashes reported in 2022, which deteriorated to 1 death per 33 crashes in 2025. This spurred a major NHTSA probe of nearly 3 million Teslas last year, and is likely what’s keeping Tesla from doing more driverless robotaxi rides in Texas.

• Heavy regulatory scrutiny of FSD: While Musk and his fans insist that Tesla has “solved” FSD, an in-depth probe by NHTSA is unearthing more flaws in the system. After forcing Tesla to search their databases with keywords like “Ran Red,” “Wrong Way,” “Double Yellow”, etc., a whopping 8,313 potential incidents turned up (cumulative FSD crashes are only 1,181).

• How Musk’s data manipulation got Tesla into hot water: Tesla is notorious for fudging its Autopilot and FSD “safety” data (famous example here), and hiding whatever shows that its self-driving software is actually dangerous. This finally caught up to Musk last August, as NHTSA launched an investigation into Tesla’s crash reporting methods, which have been as late as 5 months in reporting FSD crashes involving fatalities. All of this looks bad if you’re trying to launch a driverless ride-hailing service.

• The Cybercab launch is meaningless without FSD: Musk has hyped the Cybercab—which is so optimized for self-driving that it has no steering wheel or gas/brake pedals—and investors are waiting with bated breath for its scheduled launch next month. But, without FSD, the Cybercab is useless, and even if it had a steering wheel and pedals, few would buy the car, as only sports car connoisseurs buy cars with 2 seats. Moreover, the Texas DMV won’t be issuing AV permits until May, so there’s a lot of upcoming disappointments for Tesla bulls.

Read more

• Trouble in China: The headline news for Tesla yesterday was that domestic sales in China plummeted 45% YoY to only 18,485 cars in January 2026. While overall shipments from China rose 9% YoY, this was driven entirely by exports, which surged 113% YoY last month.

• Tesla’s exports have peaked in both volume and profitability: Exports from China outweighing domestic sales there would normally be a tailwind for margins (see Figure 1), as Tesla’s China prices are so low they approach break-even levels. Historically, exports “raked in the cash” due to generous export incentives from the Chinese government, which made Tesla’s China exports overseas, where prices are much higher, hugely profitable. However, this equation has been turned upside down.

• Profit margins on exports just shrank by 4 percentage points: Tesla’s aging lineup is colliding with China’s initiative to wean EV makers off subsidies. The VAT rebate for exports from China was recently cut from 13% to 9%, which led to a $1,240 rise in costs per unit. Furthermore, Tesla’s 2025 annual report showed US profits outweighing “foreign” pretax profits for the first time—a staggering shift. While analysts obsess over things like the bill of materials for Optimus, etc., they’re missing this major “domestic vs. foreign” profit flip (see Figure 2), which needs a lot of explaining, given the low profitability of Tesla’s US operations.

• Despite recent “refreshes,” Tesla is forced to cut prices: Model Y sales in China fell from the best-selling electric SUV in December 2025 to 20th place in January 2026. Meanwhile, a Chinese smartphone maker called “Xiaomi”, outsold Tesla’s Model Y by over 2x (39,002 units vs the Model Y sales of 16,845). This comes after only 12 months since Tesla conducted a costly “refresh” of the Model Y, which merely changed the head and tail lights, along with a few interior upgrades. This doesn’t pass as a “full makeover”, which all carmakers do every 2 to 4 years. Musk scrapped the $25,000 entry-level Tesla Model 2 (which would be selling like hotcakes in the US and other markets that suffer from car affordability), but he opted for launching the Cybercab, which will only sell if FSD is approved as a Level 4 self-driving system.

• Giga Shanghai’s shrinking export markets: Exports are shrinking as a percentage of total production because Tesla has the oldest lineup in the industry. To move these aging models, Tesla has been forced to cut prices heavily in its primary export hubs. Figure 3 illustrates this: in every market that grew over the last two years, Tesla sacrificed price (e.g., Korea and Japan). Oceania serves as a cautionary tale; despite price cuts, 2025 sales of 31,612 units were 31% below the 2023 peak, and January 2025 sales were down by 36% YoY.

• Korea is Giga Shanghai’s primary “channel stuffing” region: Local media now describes South Korea as Tesla’s “clearance outlet.” On the final day of 2025, Tesla slashed Model Y prices in Korea by 21% ($9,258) and Model 3 prices by 15% ($5,403). Remarkably, the Model Y is now 5% ($1,057) cheaper in Korea than in China. With 2025 sales in Korea already 610% above 2021 levels and BYD entering the fray, further growth for Tesla in Korea appears unlikely.

• Thailand registrations in January exceeded all of 2025: Tesla sold 5,800 vehicles in Thailand last month, surpassing its total 2025 deliveries of 5,049. This was a “pull-forward” event caused by the tightening of Thai EV subsidies on February 1, 2026, which effectively raised Tesla’s average price by $8,000. Tesla is now sitting on massive inventory in Thailand. While they introduced “Standard” (lower-spec) versions on January 17th to lower the entry price, they still face an 8% excise tax disadvantage compared to BYD, which produces locally.

• Cutting prices as raw materials spike: Battery costs are rising sharply. The price of Lithium Hydroxide (US models) and LFP materials (all other models sold outside of the US) has increased the cost per Model Y by roughly $6,765 in the US and by at least $838 in markets outside of the US. While Chinese rivals can offset these costs with massive volume growth, Tesla’s dropping volumes suggest it will likely face losses and significant cash burn in 2026.

Read more

This is not investment advice, but I’m shorting any spikes in Tesla’s stock on the back of “merger” news with SpaceX (no synergies), or xAI (even less synergies) from here on.

The rumors of a possible Tesla + SpaceX merger after horrible earnings and guidance at Tesla last week, and yesterday’s announcement of a SpaceX and xAI merger, are clear signs of desperation by Musk.

Tesla shareholders already launched a class action against Musk and the Tesla Board over Musk’s creation of xAI outside of Tesla, rather than within the carmaker (Musk diverted key talent, resources, and hard-to-get GPU orders from Tesla to xAI in late 2023).

After Musk, who has a 75%+ super-majority voting power at SpaceX, got the rocket company to acquire xAI—the worst AI among the major names—it’s only a matter of time for SpaceX shareholders to launch lawsuits against Musk and SpaceX.

And this is all because Musk is now feeling the walls close in on him more quickly than ever before. His plan to do this was always in place, but expedited after the worse-than-expected outlook for Tesla and last weekend’s disclosure that he reveled with Jeffrey Epstein in his twisted ideas of “fun”.

Figure 1 below shows earnings projections (based mostly on leaked numbers from SpaceX).

Figure 2 below shows Musk’s stakes in Tesla, SpaceX, and xAI at current valuations.

Figure 3 below shows the combined 3 entities of Tesla, SpaceX, and xAI valuations on a 2028 combined basis (note that Tesla is near bankruptcy).

Trouble in Doggie Land

In 2016, when Musk’s SolarCity faced bankruptcy, he proposed that Tesla take over the defunct solar energy company. I immediately shorted more Tesla stock, thinking its valuations would become even less attractive if it absorbed all of SolarCity’s debt and liabilities.

Musk held a demonstration of a new “solar roof tile” that SolarCity had developed, and showed a demonstration on the Hollywood set of the popular series, Desperate Housewives, of how nice the tiles would look.

The synergies between solar tiles and EVs, Musk explained, were that everyone could buy Tesla solar tiles and charge their Tesla EVs at home for free. Not only did shareholders vote in favor of the acquisition, but Tesla's stock had a relief rally that quickly squeezed me out of my short position.

These were some of the fraudulent behind-the-scenes ongoings that came out 6 years later during 2 shareholders’ lawsuits in the Delaware Court of Chancery:

• Fake solar tiles: The solar tiles that Musk displayed that day on the set of Desperate Housewives (cover picture above) were non-functional.

• Fake SolarCity numbers: Immediately after the deal closed, Ernst & Young discovered that SolarCity’s 2017 financial projections had "omitted" 2 major solar bond payments and were short of cash by $169 million. Plaintiffs argued this proved the company was "worthless" and that the $2.6 billion price tag was a gift to Musk and his family (who were major SolarCity shareholders).

• The Board conflict: It was revealed that 5 out of 7 Tesla board members had direct ties to SolarCity or Musk’s other ventures (SpaceX), meaning the "independent" review of the deal was effectively nonexistent.

• Valuation manipulation: Tesla’s merger advisor, Evercore, was pressured into “backing into” a valuation that justified the $2.6 billion price, which was based on wildly optimistic SolarCity projections.

Twitter bailed out by xAI last year: After Musk turned Twitter into an ultra-right, white supremacist social media site (as well as telling advertisers to “GFY” if they didn’t like it), xAI absorbed Twitter for the same equity value at which original investors bought it for. Fidelity had written down its investment in Twitter by 79% before it was acquired by xAI at Fidelity’s original purchase price. This was possible somehow due to “peak Musk” last March, which got investors so confident of his proximity to Trump that much of Twitter’s high-yield debt was bought out at par value.

SpaceX bails out Musk’s cash-incinerating AI startup: Yesterday, SpaceX announced that it is merging with xAI Holdings Corp. (the entity owning xAI and Twitter) in a share swap that valued xAI Holdings Corp. (“xAI” from here) at $250 billion and SpaceX at $1 trillion. This came out just after the Justice Department (DOJ) released another tranche of Epstein files, in which Musk showed up over 1,700 times.

More embarrassingly, it showed Musk eagerly asking if he could come to Epstein’s wildest parties on his infamous private island in late 2013 (5 years after Epstein became a convicted child molester), and showed that Musk replied “Yes” to an email from Epstein, saying, “Will you come to caribean [sic] this xmas? woody allen [sic] with me, you might enjoy”.

Tesla’s warning last week of 2026 cash burn: While talk of a $1.5 trillion SpaceX IPO had been around since late last year, after Tesla’s Q4 2025 earnings call last Wednesday, rumors of a Tesla + SpaceX merger began to swirl after Tesla shares fell due to its CFO guiding for 2026 capex in “excess of $20 billion” (147% YoY and 1.4x Tesla’s 2025 operating cash flow of $14.7 billion).

The SpaceX numbers being bandied around are said to be so good that if its IPO raised $50 billion at a $1.5 trillion valuation, it would be stupid not to own Tesla shares, as SpaceX’s balance sheet could more than cover “in excess of $20 billion” of Tesla’s capex needs for robotaxis, AI compute power, maintenance capex, and more, for several years (I doubt it).

Many big funds were bullish on the potential Tesla + SpaceX merger last week, as SpaceX is thought to have fatter free cash flow than Tesla from here (which isn’t saying much). But Musk poured cold water on those hopes yesterday by announcing that SpaceX would acquire xAI at a 1 to 0.146 share merger ratio, valuing SpaceX at $1 trillion and xAI at $250 billion.

This diluted SpaceX shareholders by at least $250 billion, given that SpaceX, as of Monday, was still set to IPO this June at $1.5 trillion, and it increased xAI’s valuation by 8.6% from a $230 billion valuation on the back of a January 16th funding round. As I write this note, French authorities have raided X’s Paris office on multiple charges, including:

• Complicity in child abuse: Authorities are investigating the platform’s alleged failure to prevent the possession and distribution of child sexual abuse material (CSAM).

• Sexual deep fakes: The raid follows reports that X’s AI chatbot, Grok, was being used to generate nonconsensual, sexually explicit “deepfake” images of real women and children.

• Holocaust denial: Under French law, denying crimes against humanity (such as the Holocaust) is a crime. Grok reportedly generated posts denying historical facts about the Holocaust, which triggered immediate legal scrutiny.

• Biased algorithms & data extraction: Prosecutors are looking into “fraudulent data extraction” by organized groups and whether X’s recommendation algorithms were intentionally manipulated to distort public data processing.

Potential Fines and Legal Fallout

X is facing pressure from both French criminal law and the European Union’s Digital Services Act (DSA). The UK and Spain are looking at similar charges for X. The financial stakes are historic:

1. EU/DSA Fines: The European Commission already hit X with a €120 million fine in December 2025 for “deceptive design” and transparency failures. Under the DSA, repeated or systemic failures to moderate illegal content (like CSAM or deepfakes) can lead to fines of up to 6% of the company’s total global annual turnover ($3.2 billion in 2025, so a fine of around $192 million).

2. French Criminal Fines: If found guilty of “complicity” in the spread of illegal content under the French Penal Code, the company could face multi-million euro fines per instance, though criminal penalties for executives are often the focus in such raids.

3. Summons for Musk: In a rare move, Paris prosecutors have officially summoned Elon Musk and former CEO Linda Yaccarino for “voluntary interviews” scheduled for April 20, 2026.

Projections Are Ridiculous & Cut SpaceX by 33%

Before the French prosecutor raided Twitter’s Paris office, SpaceX shareholders were already irked by the fact that Musk merged xAI into the company. Looking at Figure 1 below shows why: xAI, even when using rosy forecasts leaked from pitch books for bond investors, burned around $14 billion of cash in 2025, more than wiping out SpaceX’s $2.0 billion in estimated FCF.

While the numbers go parabolic as we approach 2030, keep in mind that the 2030 FCF numbers for SpaceX were 68% higher at $42 billion yesterday, after the SpaceX CFO held a conference call with potential investors, while the xAI 2030 FCF numbers suddenly spiked by 87% to $28 billion. I used the pre-announcement projections gathered from numbers floating around before yesterday’s rushed release.

Read more