The buyback was the easy part. OpenAI completed a $7 billion buyback of employee shares, purchasing stock from current and former staff at a valuation of $852 billion, Bloomberg reported Monday. The harder part begins now, because that price tag was just stamped onto the ledger days before the company must hand every institutional investor on the planet its first look at independently verified financials. The public S-1 prospectus is expected on SEC EDGAR in mid-to-late August 2026, roughly 15 days before any roadshow. The sequence is tight, the stakes are not abstract, and the gap between what the market thinks it knows and what the auditors actually found is the most important spread in finance right now.
What makes the timing of the buyback genuinely unusual is the mechanism. Rather than lining up outside investors to absorb the secondary stock, OpenAI used its own balance sheet. That is a deliberate signal and a deliberate silence simultaneously. Analysts point out that implementing buybacks with self-owned funds rather than external financing helps streamline the shareholder registry and reduce the complexity of the equity structure before going public, while avoiding prices set by external investors from serving as a reference for subsequent IPO pricing, thereby retaining greater flexibility during the official bookbuilding process. In plain language: by keeping outsiders out of the transaction, OpenAI kept control of the only valuation number that will matter when roadshow bankers walk into conference rooms next month. The result is the first flat valuation print in a long run of escalating tenders. OpenAI moved from a $157 billion valuation in October 2024, to $300 billion in March 2025, to $500 billion via an employee tender in October 2025 that Thrive Capital, SoftBank, Dragoneer, MGX and T. Rowe Price bought into, and then to $852 billion after the March 2026 round closed. Flat is not a retreat; it is a calculated pause before the public market sets the next number.
The prospectus that drops into EDGAR will be unlike anything investors have seen from an AI company at this scale, because it will carry numbers that are actually audited. Every figure circulating in analyst decks right now is a management estimate or a leaked projection. The public S-1 amendment filed before the roadshow will be the first look at audited financials, full risk factors, and revised Microsoft terms. The Microsoft piece alone is consequential: the company faces severe cash burn, driven by massive R&D, inference costs, and $17.2 billion in Microsoft service fees. According to disclosures by The Information, OpenAI consumed $3.7 billion in cash in the first quarter of 2026, exceeding half of its $5.7 billion revenue for the same period. What the auditors say about how that burn rate is trending, and whether the unit economics of inference are improving fast enough to matter, will arrive simultaneously on every institutional desk the moment the S-1 goes live, with no friendly pre-briefing and no off-the-record framing.
The bulls have a case. On August 13, OpenAI's annualized revenue surpassed $40 billion, doubling from the end of 2025. Weekly active users for ChatGPT stood at approximately 905 million, while the number of enterprise customers doubled year-over-year. The improvement levers are well understood, including cheaper inference hardware, model distillation, batch optimization, and custom silicon, and a disclosure that shows unit economics turning positive by 2028 prices the deal favorably. The skeptics are equally prepared. FutureSearch estimates 2026 GAAP losses at $25 to 26 billion, approximately 80 percent higher than the $14 billion non-GAAP figure most headlines cite, and when the audited S-1 arrives, the GAAP number will be the one public investors see first; if it is significantly above market expectations, demand could price the offering below $1 trillion. Sam Altman is reportedly treating a sub-trillion price as a non-starter, which creates an unusual dynamic: according to The New York Times, OpenAI's IPO timeline has been delayed from the initially expected fall of this year to 2027 if the numbers do not support the ambition.
The competitive context sharpens the pressure further. Investment bankers have advised both OpenAI and Anthropic that an early mover advantage is at stake; the first to list would set the terms for how investors categorize the AI sector and gain access to enormous amounts of capital looking for an entry point. Whoever lands the first frontier-AI IPO sets the comp for everyone else: a successful debut at $1 trillion builds the runway for Anthropic to price at parity, while a flat debut at $850 billion forces Anthropic to discount. The governance structure adds its own complication: the combination of a CEO with no equity, a nonprofit foundation holding a meaningful stake, and a complex public benefit corporation structure is unprecedented for a US tech listing at this scale. None of that kills a deal. All of it makes the S-1 the most closely parsed document the AI industry will have produced for public consumption.
The $852 billion buyback price is not a valuation; it is a hypothesis, and the audited S-1, arriving any day on EDGAR, is the first empirical test of whether the market agrees.