On 2026-06-25 the New York Times reported that OpenAI is leaning toward a 2027 IPO rather than the late-2026 debut Sam Altman has pushed for, with CFO Sarah Friar telling colleagues the company isn’t ready for public markets. The reversal lands while Anthropic’s confidential S-1 is already in SEC review. It inverts the assumed order of the two-horse race: the first audited template for how Wall Street prices a frontier-model lab now looks likely to come from Anthropic’s numbers, not OpenAI’s.
Why is OpenAI leaning toward 2027?
OpenAI’s leadership is split on timing, with Friar arguing the company should wait and Altman still pressing for an earlier listing. People involved in the company’s deliberations describe the delay as a symptom of “the unclear paths of A.I. titans,” according to the Times reporting summarized by NextBigFuture: the longer a lab waits, the more it has to prove about a business model no public market has priced yet.
The mechanics matter more than the headline. Axios reported that OpenAI confidentially filed IPO paperwork with the SEC on 2026-06-08. A confidential filing is not a commitment to list, though. It starts the SEC review clock, and the actual public debut stays a separate, movable decision. NextBigFuture references OpenAI’s earlier plans for a late-2026 debut; the Times now says that timeline is slipping to 2027.
That distinction is the whole story. Both labs have confidential filings in review. Anthropic submitted its draft on 2026-06-01, OpenAI a week later. The race the analyst press keeps score on is who actually prices and lists first, and a 2027 lean hands that position to Anthropic.
What is already in Anthropic’s SEC filing?
Anthropic confidentially submitted a draft Form S-1 on 2026-06-01, with no share count, price range, ticker, exchange, or IPO date set. The filing gives the company the option to go public after SEC review. Two analyst outlets, TechStackIPO and Outgrave, characterize it as the first AI lab to start the SEC clock. That is analyst framing, not Anthropic’s own language; the company’s public posture has not claimed the distinction.
The filing rests on a $65 billion Series H that closed 2026-05-28, four days before the submission, at a roughly $965 billion post-money valuation, per Outgrave, led by Altimeter, Dragoneer, Greenoaks and Sequoia, with Capital Group, Coatue, GIC, ICONIQ and XN as co-leads. Outgrave reports Anthropic’s run-rate revenue crossed $47 billion earlier in May, up from $14 billion in February and roughly $10 billion a year earlier. Those numbers travel through secondary reporting rather than an audited filing, and run-rate is annualized, so it is not directly comparable to OpenAI’s quarterly revenue.
Even taken at face value, $965 billion against roughly $47 billion in annualized revenue implies about a 20x multiple, against the 8 to 15x range typical for high-growth public SaaS. Outgrave draws the same comparison. That gap is the bet the market will have to price.
Why does whoever lists first anchor the AI-lab multiple?
A public comparable anchors how investors underwrite the next lab. Once Anthropic’s prospectus publishes audited revenue, gross margin, compute obligations and customer concentration, every private AI lab negotiating its next round does so against a number that did not exist before. A 2027 OpenAI delay hands the anchor-setting role to Anthropic, and the cost of going second runs both ways. If the market flinches at compute-as-COGS and Anthropic’s multiple prints low, OpenAI inherits the discount it had no hand in earning. If it prints high, OpenAI can ride the momentum but loses the narrative of setting the ceiling itself.
Outgrave flags several reasons Anthropic may file first and cleaner: an enterprise-first go-to-market against OpenAI’s consumer weighting, a cleaner PBC structure, a faster product cadence on the Opus 4 family, and fewer open governance questions. Reuters reported in April that Anthropic’s annualized run-rate is growing faster than OpenAI’s at comparable stages. Those are secondary characterizations, but they shape how the analyst class will frame whichever prospectus lands first.
The under-covered consequence is downstream. A published multiple does not just move the two incumbents. It re-prices every lab still raising privately, because the first audited frontier-model comparable becomes the reference point term sheets get written against. A lab that might have commanded 30x on a private deck in 2025 suddenly has to justify itself against a printed number, and that number is set by whichever prospectus the SEC clears first.
What will a frontier-lab prospectus force into the open?
A prospectus forces onto the balance sheet the things private decks can hand-wave. Frontier-model economics turn on three lines that have never appeared in a public AI-lab filing: compute commitments booked as long-dated obligations, hyperscaler equity and revenue concentration, and gross margin after inference cost. The last one is the crux. A lab can book billions in subscription revenue and still bleed margin if inference eats the take, and public investors have no audited baseline for where that line sits.
OpenAI’s reported stack gives a sense of the scale. Per secondary reporting, the company generated roughly $5.7 billion in Q1 2026 revenue against about $3.7 billion in cash burn that quarter, with projected losses near $14 billion for the year. Its infrastructure includes the $500 billion Stargate program, a more-than-$300 billion Oracle capacity arrangement, a $250 billion Azure deal running through 2032, and roughly $138 billion with AWS. Those figures are not audited and should be read as directional until a filing confirms them.
The structural facts are firmer. Under the 2025 PBC restructuring, Microsoft holds 27%, the OpenAI Foundation holds 26%, and employees and investors hold 47%. OpenAI agreed to purchase $250 billion of Azure services from Microsoft.
What does a 2027 delay keep off OpenAI’s disclosure?
For OpenAI, waiting a year keeps the uncomfortable disclosures private. The PBC restructuring mechanics, the Microsoft commercial terms, the Stargate, Oracle and AWS obligations all stay off the public docket while the listing slips. That is the disclosure asymmetry the horse-race coverage mostly misses: a delayed filing means the market’s first audited read on frontier-model unit economics gets set by Anthropic, and OpenAI’s specific terms stay hidden for another year.
The valuation figures are where the speculation is loudest, and where the sources disagree. NextBigFuture’s summary of the Times reporting puts OpenAI’s last private raise at about $730 billion and says Altman is pushing for a $1 trillion IPO valuation; StartupFortune cites a valuation north of $850 billion. None of these is verified. All of them collapse to guesses until a prospectus publishes a number, and the gap between $730 billion and $1 trillion is wide enough that the anchoring game cuts both ways.
What does SpaceX’s debut say about pricing risk?
SpaceX’s first two weeks as a public company are the live cautionary tale for anything harder to price than a launch operator. Per StartupFortune, the stock priced at $135 a share, opened at $150 on 2026-06-12, peaked at $225.64 on 2026-06-16, then fell to $154.54 by 2026-06-24. The peak sat roughly 67% above the offering price; the stock then fell about 31% over the following week. That is a two-week swing on a company with comparatively clean physical-asset economics.
The implied read for OpenAI is direct. A frontier lab carrying capped-profit overhang, contingent commercial obligations to Microsoft, and multi-hundred-billion-dollar compute commitments is harder to price than rockets, not easier. Retail volatility on that mix is the risk Friar’s caution is pointing at, and it is a reasonable argument for letting a cleaner filing go first to set the sector’s tone.
What should operators and enterprise buyers watch next?
The filing order, the compute-as-COGS line, and the first published multiple are the three signals that will reshape how the sector is priced. Which lab’s prospectus lands first sets the comparable everything else gets read against. The gross-margin-after-inference disclosure will be the most-parsed number in whichever filing arrives, because it is the line no public AI lab has had to publish before. Hyperscaler concentration and the Microsoft commercial terms will reveal how much of frontier-model revenue is effectively pre-committed to a cloud partner. And the first published multiple will re-price private rounds across the sector, not just for the two incumbents.
Until an actual S-1 publishes audited figures, the valuations, the revenue numbers, and the timing all stay speculative. Treat any single leaked figure as a data point about positioning, not a fact about the business.
Frequently Asked Questions
Beyond the Azure purchase commitment, what other financial obligation to Microsoft would appear in an OpenAI S-1?
OpenAI’s PBC restructuring ties the company to sharing 20% of its revenue with Microsoft until an independent panel declares AGI has been achieved. That contingency puts an open-ended cash-flow obligation on the income statement whose duration is undetermined, depending on a judgment call no public market has ever had to underwrite before.
Did OpenAI’s CEO and CFO align on IPO readiness before the Times story broke?
No. The Information, cited in secondary reporting, noted that Altman excluded CFO Sarah Friar from some conversations about OpenAI’s financial plans before the Times reported her warning. A CFO kept outside certain financial discussions who then tells colleagues the company is not ready for public markets is a more pointed governance signal than a routine executive disagreement over calendar timing.
Why does an enterprise-weighted revenue mix produce a tighter valuation range than a consumer subscription base?
Enterprise contracts typically carry multi-year terms, defined renewal cycles, and predictable churn rates that underwriters can model with standard metrics like net revenue retention. Consumer subscriptions are month-to-month with cancellation risk that spikes when a competitor publishes a free-tier equivalent. A high enterprise percentage lets underwriters narrow the revenue quality band; OpenAI’s consumer weighting adds a variable Anthropic’s prospectus will not need to defend in the same way.
What should enterprise buyers review in their existing lab contracts before either S-1 publishes hyperscaler concentration numbers?
Buyers should pull their exit clauses and force-majeure terms before the first prospectus prints audited capacity-commitment figures. If a lab discloses that the large majority of its inference capacity is pre-committed to one cloud provider, any pricing renegotiation or extended outage at that provider cascades directly into the buyer’s SLAs. The S-1 will be the first chance to size that single-provider risk against an audited number rather than a vendor assurance.
