The SpaceX S-1 contains a detail that most coverage of the frontier AI IPO season has underplayed. Anthropic and OpenAI, two of the three labs heading toward public markets, are listed among the largest compute customers of xAI’s infrastructure division. They’re rivals renting compute from each other’s parent company. That’s not a footnote. It’s the clearest illustration of how structurally entangled frontier AI economics have become, and why investors evaluating these IPOs separately are asking the wrong question.
The three-lab public market moment
Three timelines are running in parallel. SpaceX filed an S-1 prospectus covering xAI’s financials, the only audited public financial disclosure among the three major frontier labs. Anthropic announced on June 1 that it has filed a confidential draft S-1 with the SEC, initiating the review process that typically precedes a public offering by six to twelve months. OpenAI has been reported to be targeting a fall 2026 public market debut, with Goldman Sachs and Morgan Stanley reportedly positioned as potential lead underwriters per Fortune’s reporting, though OpenAI hasn’t confirmed that timeline.
Three labs. Three sets of expectations. One audited data source between them.
What the xAI S-1 actually shows
The numbers in the SpaceX S-1 are the most important data in this piece, because they’re the only numbers not sourced to a vendor’s communications team.
Per the S-1 prospectus: xAI generated $3.2 billion in revenue in 2025. Its operating loss was $6.36 billion. Capital expenditures reached $12.7 billion for full-year 2025, then accelerated to $7.7 billion in Q1 2026 alone, a quarterly run rate that, if sustained, puts 2026 CapEx above $30 billion. Cash burn in Q1 2026 was approximately $2.5 billion.
These figures have been in the public record since at least late May, and they reveal something specific: xAI’s economics don’t work as a standalone business at current scale. The operating loss ratio, roughly $2 in losses per $1 in revenue in 2025, is characteristic of a lab in aggressive infrastructure build-out, not a maturing software business.
What makes this sustainable, at least on paper, is Starlink. The same S-1 shows Starlink generated $11.4 billion in revenue and $4.4 billion in operating profit in 2025. Starlink’s margin is subsidizing xAI’s burn. That’s the cross-subsidy model: a profitable adjacent business funds the losses of the AI lab while it scales toward (theoretically) better economics.
No other frontier lab has disclosed a comparable cross-subsidy structure in audited form.
Analysis
The xAI S-1 discloses Anthropic and OpenAI as among the largest compute customers of SpaceX's AI infrastructure division. Investors evaluating any one of these three IPOs are implicitly exposed to the others, competitive, supplier, and capital relationships run on the same infrastructure.
Frontier Lab IPO Readiness, Stakeholder Positions
Evidence
The transparency asymmetry
Anthropic has reported an annualized revenue run rate of $47 billion, though this is a gross figure before hyperscaler revenue-share retrocessions, and the company’s own communications acknowledge the gap. The distinction between gross ARR and net revenue matters significantly: prior reporting has documented that Anthropic’s net revenue figure sits well below its headline ARR, reflecting the cost of the compute agreements with Amazon and Google that underpin its infrastructure. Anthropic’s most recent funding round assigned a post-money valuation of $965 billion, a private-market figure, not a public market price. The confidential S-1 filing means the actual financials remain under SEC review, not public.
OpenAI’s situation is similar in form if not in detail. A fall 2026 debut has been reported; internal deliberations over the IPO timeline, including reported CFO-level concerns about readiness, have appeared in prior coverage. Neither the revenue mix, operating structure, nor capital requirements are publicly disclosed in audited form.
The asymmetry is stark. Investors evaluating the Anthropic and OpenAI IPOs will do so with vendor-reported revenue figures, private-market valuations as reference points, and no audited operating loss data. The xAI S-1 exists precisely because SpaceX, as the filing entity, had the regulatory obligation to disclose. Anthropic and OpenAI, as private companies filing confidential S-1s, won’t release public financials until their registration statements are declared effective.
What Berkshire’s commitment adds
Alphabet’s June 2 announcement of an $80 billion capital raise with Berkshire Hathaway anchoring at $10 billion is not directly an AI lab IPO event. But it belongs in this analysis. Berkshire’s historical posture toward technology capital expenditure at scale was skepticism. That’s shifted. While Berkshire’s investment is in Alphabet’s infrastructure, not directly in the lab IPO process, the commitment represents a meaningful data point about how generalist value capital is now engaging with AI infrastructure at anchor scale.
The infrastructure these labs run on is being funded by a broadening set of institutional capital. That’s the context investors need when evaluating whether xAI’s cross-subsidy model is idiosyncratic or indicative. If Berkshire is willing to anchor Alphabet’s infrastructure at $10 billion, the thesis that AI infrastructure yields are eventual and worth the wait has cleared a credibility threshold it didn’t have two years ago.
Three questions the IPO process should answer, and current disclosures don’t
The xAI S-1 is useful not just for what it confirms, but for what it reveals as the right questions to ask of the others.
First: net revenue versus gross ARR. Anthropic’s $47 billion run-rate figure and its actual net revenue after hyperscaler retrocessions are materially different. The S-1 will need to address this; the confidential filing period won’t. Investors should model the gap before assuming ARR is a clean revenue proxy.
What to Watch
Warning
Three labs, three IPO timelines, one audited data set. The xAI S-1 is a reference point, not a blueprint, Starlink's cross-subsidy is structurally specific to SpaceX and doesn't translate to labs whose compute dependencies run through the same hyperscalers they compete against.
Second: cross-subsidy sustainability. Starlink’s model works because satellite internet is a distinct and profitable business. Anthropic’s AWS partnership and OpenAI’s Azure dependency are compute arrangements, customer relationships with the hyperscalers, not independent profit centers. The hyperscaler relationship is both a distribution asset and a margin constraint. What cross-subsidizes the loss structure when the partnership terms reprice?
Third: compute customer concentration. The xAI S-1 discloses that Anthropic and OpenAI are among its largest compute customers. That means competitive relationships and supplier relationships are running on the same infrastructure. Any material change to those arrangements, pricing, availability, or strategic redirection, is a risk factor for all three labs simultaneously. Investors in any one of these IPOs are implicitly exposed to the others.
What to watch
The Anthropic confidential S-1 begins its SEC review period now. The standard review window runs six to twelve months before a registration statement can be declared effective. A 2027 public debut is more likely than a late 2026 listing given that timeline. OpenAI’s reported fall 2026 target is aggressive, watch for any public statement from OpenAI itself, which hasn’t confirmed the timeline attributed to it in trade press.
The xAI roadshow, when it begins, will be the first live market test of investor appetite for a frontier lab at this loss ratio and CapEx rate. How that prices will set the reference frame for Anthropic and OpenAI whether they want it to or not.
Don’t bet on all three reaching public markets on the timelines currently circulating. Regulatory review periods, market conditions, and the Q2 earnings cycle, which will produce the first hard data on hyperscaler AI infrastructure returns, all create windows where one or more of these timelines slip. The lab that goes last will have the most pricing data and the most scrutiny.