SpaceX IPO update
The SpaceX S-1 has still not dropped and the valuation is still sitting at $2 trillion on narrative alone.
This is a follow-up to my first SpaceX IPO piece, which covered the valuation structure, the NASDAQ rule change, the Morgan Stanley conflict of interest, and the Tesla acquisition thesis. If you have not read that one, you can find it here.
This update covers three developments since then that change parts of the picture. The S-1 still has not dropped yet so everything before it is still a pitch.
If you want to follow the portfolio behind the analysis, you can find me here.
Three things that have changed
The SpaceX IPO narrative has moved considerably since my first piece. The valuation has held at $2 trillion. The filing has not dropped. But three specific developments have added new analytical layers that are worth mapping before the S-1 arrives.
The Google deal and what it signals
Google is in talks with SpaceX to secure preferential rocket launch access. Google owns approximately 6.1% of SpaceX, making it one of the earliest and largest outside investors. The deal centres on Project Suncatcher, Google’s internal initiative to put its TPU chips into orbital data centers. Google has already done the foundational work: its TPUs have been tested in space and survived 15 kilrads of radiation in testing, roughly three times the expected five-year LEO exposure at small scale. Most commercial space-rated systems targeting decade-long reliability aim for 50 to 100 kilrads, so this is a proof of survival rather than a proof of readiness. It has also partnered with Planet Labs, providing two prototype satellites to evaluate whether TPUs can operate at commercial scale in orbit.
Google has been working toward orbital data centers for years. Securing preferential launch access before the IPO locks in its position as anchor tenant if the Starship cost curve plays out. The deal also strengthens the IPO narrative in the short term, which benefits Google's existing stake regardless of whether the long-term space thesis works out.
The space data center reality check
The orbital data center thesis deserves more scrutiny than it typically receives. There are three unsolved engineering problems between the current state and the commercial reality the $2 trillion valuation assumes.
The first is launch cost. Current SpaceX Falcon 9 costs run between $250 and $600 per kilogram to orbit. SpaceX is targeting $50 per kilogram at Starship scale, which would require full routine reusability at a level not yet demonstrated in regular operations. The commercial viability threshold for space data centers is estimated at below $200 per kilogram. The gap between current reality and the threshold is significant and requires Starship to perform at a level it has not yet sustained.
The second is radiation hardening. Google’s TPU testing is genuinely encouraging. Surviving 15 kilrads at small scale is a meaningful proof of concept, roughly three times the expected five-year LEO exposure at that scale. However most commercial space-rated systems targeting decade-long infrastructure reliability aim for 50 to 100 kilrads. Google’s TPUs are surviving current testing conditions. They are not yet proven reliable for the kind of decade-long infrastructure play that the orbital data center thesis requires. Scaling from a test chip to a gigawatt data center also introduces compounding failure risks that have not been tested at commercial scale. High bandwidth memory, which modern AI chips depend on heavily, is particularly vulnerable to cosmic radiation bit-flip events. The components that make current AI chips fast on Earth are the same components most likely to fail in space.
The third is heat dissipation. Space has theoretical cooling advantages because the ambient temperature can reach near absolute zero. In practice, heat dissipation in a vacuum requires radiating the heat away from the chips through physical surface area rather than airflow. Current estimates suggest approximately four tennis courts of radiating surface per megawatt of compute. At gigawatt data center scale that becomes an enormous physical engineering challenge with no established solution at commercial scale. This is the most underreported technical risk in the entire space data center narrative.
None of these problems are fatal but all of them require time and capital to solve. The $2 trillion valuation is pricing the solved version. The engineering reality is somewhere between the current state and that solved version, on a timeline that nobody outside SpaceX can reliably estimate.
The XAI signal
XAI, which SpaceX reportedly acquired in an all-stock deal earlier this year (a transaction that remains unconfirmed until the S-1 is public), has signed an agreement to supply Anthropic with 300 megawatts of compute capacity across 220,000 Nvidia GPUs by end of May. This is the detail that deserves the most attention of anything in this update. To be clear: this is the terrestrial Colossus facility in Memphis. These GPUs are on the ground, not in orbit.
XAI is supplying compute capacity to its direct competitor. Anthropic usage at enterprise companies has reportedly quadrupled. On every observable usage metric, enterprise adoption, coding tools, model performance benchmarks, Anthropic is currently ahead of XAI’s Grok. The Cursor arrangement, widely described as a $60 billion deal, is more precisely a $60 billion option. SpaceX can walk away for a $10 billion partnership fee. One framing is an acquisition. The other is a marketing commitment with an exit clause. The XAI deal with Anthropic is a revenue patch on a business burning approximately $1 billion per month. It confirms that XAI’s revenue reality is significantly weaker than its narrative implies, and that the pressure to generate cash before the IPO is real enough to supply the competition.
In my first piece I noted that the $8 billion EBITDA figure circulating in secondary markets predated the XAI acquisition. The Anthropic deal adds another data point to that picture: XAI’s losses are large enough that the business is willing to generate revenue from any source available, including direct competitors. The S-1 will show what that actually looks like in audited numbers. Until then, the XAI burn rate is one of the three most important unknowns in the entire SpaceX valuation.
What this adds to the Tesla thesis
In my first piece I laid out the Tesla acquisition thesis in full. This update adds two pieces of evidence without changing its speculative character.
Every deal made before the IPO has been about derisking SpaceX and improving what the S-1 will show. The Cursor option gave XAI a customer. The Anthropic deal gives XAI revenue. Both reduce the losses that will appear in the filing. A cleaner S-1 supports a higher IPO valuation. A higher valuation on a 6 to 7% float creates the squeeze mechanics the thesis depends on.
The governance fortress
On May 14th, CalPERS and the NYC Comptroller issued a joint letter formally objecting to the SpaceX IPO structure. These are institutions managing almost $1 trillion in assets. Their objections are very specific: Elon controls 79% of votes while owning approximately 42% of equity through super-voting shares, and the reported S-1 draft includes mandatory arbitration for all shareholder disputes, which blocks class-action lawsuits. They are calling this structure novel and extreme and demanding a one-share one-vote sunset before listing.
The mandatory arbitration clause is the detail that connects most directly to the Tesla thesis. It removes the legal mechanism shareholders would normally use to challenge how management deploys the balance sheet. If Musk cannot be removed from SpaceX and shareholders cannot sue as a class, he can use SpaceX’s capital with total autonomy after the IPO. That is precisely the condition the leveraged buyout of Tesla requires.
The institutional boycott scenario
If CalPERS and major pension funds boycott the IPO over governance, the low float squeeze becomes more violent. The buyers left are retail investors and sovereign wealth funds. Neither group has the same governance sensitivity as institutional asset managers. A boycott by the world’s largest pension funds does not kill the IPO, but it changes who owns it after listing and removes the institutional constituency most likely to push back on management decisions post-listing.
Watch how many major institutional names appear in the S-1 book when it drops. If the allocation is dominated by sovereign wealth and retail rather than the major pension funds and endowments, that tells you more about the post-listing dynamics than the valuation does.
Google versus the field
Alaska regulatory filings confirmed Google held 6.1% of SpaceX as of year-end 2025. Following the reported SpaceX-XAI merger in February 2026, that stake is estimated to have diluted to approximately 5.0% of the combined entity. At $2 trillion that is still worth around $100 billion. Google also has a decade of TPU development, proven radiation hardening at small scale, and a core business in cloud, advertising, and enterprise software that generates real revenue regardless of whether orbital data centers work on any particular timeline. Meta has no space plans. Microsoft has no orbital compute strategy. Amazon has Blue Origin but has not demonstrated the same launch cost trajectory. The space bet is a call option on top of a functioning profitable business. For SpaceX, the space bet is the business. That asymmetry is worth holding in mind when the S-1 arrives.
What has not changed
The S-1 has still not dropped. The Morgan Stanley conflict, the NASDAQ float waiver, the low float mechanics, and the $500 billion valuation move on narrative alone are all unchanged. The audited XAI burn rate, the real Starlink ARPU, the debt load across the ecosystem, and what the $75 billion raise is officially earmarked for are all still unknown. The governance objections from CalPERS and the NYC Comptroller are now on the record.
Everything before the filing is still a pitch.