Introduction

On Friday, May 22, 2026, OpenAI handed Wall Street a document it had been waiting years to receive. The company confidentially filed its S-1 registration statement with the Securities and Exchange Commission, officially beginning the process that could produce the largest technology initial public offering in history. Goldman Sachs, Morgan Stanley, and JPMorgan are leading the deal. The target valuation sits between $852 billion and above $1 trillion. The listing window, according to Sam Altman, is as early as September 2026. Fortune, CNBC, Reuters, Axios, and Bloomberg all confirmed the filing within hours of each other, which is what happens when the most anticipated IPO in the AI era finally starts moving.

What you cannot read yet is the actual filing. That is the point of a confidential S-1 submission. The full financial details — the revenue breakdown, the cost structure, the specific risk disclosures, the use of proceeds, the share class structure, and the terms of the unusual governance arrangement that distinguishes OpenAI from every other company that has ever attempted to go public — will not be made public until approximately 15 days before the investor roadshow begins. What is available now is a mix of public reporting, the company's own statements, and analyst estimates that together paint a picture of what the prospectus will need to explain, and what it has the luxury of keeping quiet until the last possible moment.

The Revenue Numbers That Are Genuinely Extraordinary

Start with the number that makes the trillion-dollar valuation partially defensible. OpenAI is generating approximately $2 billion per month in revenue. By March 2026, the company had reached an annualised revenue run rate of approximately $25 billion. It counts 50 million consumer subscribers to ChatGPT Plus and related products, plus 9 million business users. Enterprise contracts now account for more than 40% of revenue. The company processes 15 billion tokens per minute through its API, a figure that measures actual usage across every product and customer simultaneously.

The growth rate behind those numbers is the figure that Wall Street is most focused on. Multiple independent analyses have noted that OpenAI is growing its revenue approximately four times faster than Alphabet and Meta were growing at comparable stages of their development. That comparison is imprecise — stage comparisons across companies that operate in different eras and with different business models require significant caveats — but the directional point is accurate enough to explain why serious investors are willing to discuss a valuation above $1 trillion for a company that does not yet generate a dollar of profit. What they are pricing is not current earnings. It is the trajectory.

The Losses That Are Also Genuinely Extraordinary

Now for the other side of the ledger. OpenAI lost approximately $1.22 for every dollar it earned in the first quarter of 2026. The company is projected to lose approximately $14 billion in fiscal 2026, and its inference costs alone — the cost of running model queries across its customer base — reached $8.4 billion in 2025 and are projected to reach $14.1 billion in 2026, a 68% increase in a single year. Total operating expenses for 2026 are estimated by multiple analyst teams at somewhere between $50 billion and $57 billion, against projected revenues of approximately $29 billion to $36 billion.

The comparison that the bears reach for when they see those numbers is instructive. When Amazon built its AWS infrastructure in the 2000s, the capital was expensive up front but the marginal cost of adding a new customer was close to zero once the infrastructure was in place. A server that hosts a thousand accounts does not cost significantly more to run than a server hosting a hundred. That is the economics of infrastructure at scale. OpenAI's economics do not work that way. A GPU cluster serving a million users costs proportionally more to run than a cluster serving a hundred thousand, because each query genuinely consumes compute that must be paid for. The inference cost per token has been falling as hardware improves and model efficiency increases, but it has not fallen fast enough to close the gap between revenue and cost as volume grows.

The company does not project profitability until approximately 2029 to 2030. That is a three to four year runway during which OpenAI will need to raise capital continuously — from private investors before the IPO and from public markets after it — while operating at a substantial loss. The IPO itself is, in part, a capital raise to fund that runway.

The Governance Structure That Every Investor Needs to Understand

The single most unusual fact about the OpenAI IPO is not the valuation, the losses, or the absence of profitability until the end of the decade. It is the governance structure of the entity asking investors to buy its shares.

OpenAI's 2025 corporate restructuring converted the company into a public benefit corporation — a PBC — called OpenAI Group PBC. The original nonprofit that founded the company in 2015 was renamed the OpenAI Foundation and retained a 26% equity stake in the for-profit entity. That 26% stake is not the source of the Foundation's control over the company. The source of control is something more significant: through special governance rights held solely by the OpenAI Foundation, the Foundation appoints all members of the board of directors of OpenAI Group and can remove them at any time. Microsoft holds 27% of the equity. Employees and other investors hold the remaining 47%. Neither Microsoft, with a larger equity stake than the Foundation, nor any other shareholder holds the power to seat or remove a board member without the Foundation's consent.

What this means in practice for public market investors is that they will be purchasing shares in a company whose board is controlled by an entity that does not answer to them. The Foundation's board has nine members, eight of them independent directors and one of them Sam Altman. If the Foundation's board decides that Altman's leadership is inconsistent with the mission of beneficial AI development — which is exactly what happened in November 2023, before investors and employees forced his return — the Foundation can remove the entire board of OpenAI Group and replace it. Public shareholders can vote and object, but they cannot override the Foundation's governance rights through standard shareholder mechanisms.

The OpenAI Foundation also holds a warrant that allows it to receive additional shares if OpenAI Group's share price increases more than tenfold over fifteen years. The company's structure page describes this as positioning the Foundation to be "the single largest long-term beneficiary of OpenAI's success." That is accurate. It is also an unusual arrangement for a public company. Investors who buy into a dual-class share structure, where founders hold supervoting shares, are at least dealing with individual human beings whose economic interests are aligned with the company's market value. Investors in OpenAI Group PBC will be dealing with a nonprofit board whose primary obligation is to "the benefit of humanity broadly," as the original founding documents put it — not to maximising shareholder returns.

What Sam Altman Said About Timing

Sam Altman's public statements about the IPO timeline reveal the specific calculation that is driving the filing date. On June 8, 2026, when OpenAI confirmed publicly that it had filed for an IPO, the company added a notable qualification: it said "it could be a while because there are things we want to do that are likely easier as a private company." Altman separately told staff in an internal communication that week that he expected an IPO "within the next year" and that the faster the company develops recursive self-improvement, "the more it could be advantageous to delay an IPO."

Both statements are honest about the tension at the centre of the decision. An IPO is a capital raise that provides liquidity for early investors and employees, establishes a public market value for the company's shares, and gives OpenAI access to public equity markets for future fundraising. But a public company faces disclosure requirements, short-term earnings pressure, and shareholder activism that a private company does not. Altman's comment about things being "easier as a private company" is a genuine acknowledgement that the most ambitious AI research agenda — including whatever OpenAI means by "recursive self-improvement" — may be more tractable without the quarterly reporting cycle and investor relations obligations that come with being listed.

What moved the timeline forward despite those concerns was the legal and financial context. Elon Musk's long-running lawsuit against OpenAI, which had been a material overhang on any IPO plan, was dismissed by a jury in May 2026. The lawsuit had sought damages reportedly in the range of $79 billion to $134 billion and had raised questions about the legitimacy of the corporate restructuring that is the legal prerequisite for the offering. With the lawsuit resolved, the path was clear. The company had also closed a $122 billion funding round in April 2026 at a valuation of $852 billion, with Amazon, SoftBank, and Nvidia among the participants, establishing a private market reference point that gives the IPO valuation a grounded baseline to work from.

What the Confidential S-1 Keeps Hidden

Under SEC rules, a company that files confidentially is not required to make its prospectus public until approximately 15 days before the investor roadshow begins. OpenAI filed on May 22. If the September listing target holds, the public filing would arrive sometime in late August, meaning that investors are currently pricing and discussing a trillion-dollar offering based on public reporting and analyst estimates rather than audited financial disclosure.

What the prospectus will eventually reveal, and what is currently kept from public view, includes the specific breakdown of revenue by product and customer type; the precise cost structure, including how much compute is purchased from Microsoft's Azure versus other providers and on what terms; the detailed terms of the Microsoft partnership, which was renegotiated in April 2026 to remove Azure exclusivity and cap revenue-share payments; the governance provisions that will apply specifically to public shareholders; the risk disclosures around competition from Anthropic, Google DeepMind, Meta, and Chinese AI labs; and the specific financial projections that support the profitability timeline.

There is also a question the S-1 will need to address about the government. In June 2026, it was reported that OpenAI and the White House had been in ongoing talks for approximately a year about the possibility of a US government stake in the company. If the government holds equity in OpenAI, the implications for the company's independence, its research direction, and its international commercial relationships are significant enough to require substantial disclosure. Whether those discussions produced a formal arrangement, a memorandum of understanding, or nothing — and whether that arrangement would need to be disclosed as a material relationship in the prospectus — is a detail the confidential filing process has so far kept quiet.

What the Valuation Implies

A valuation between $852 billion and $1 trillion, against projected 2026 revenues of approximately $29 billion to $36 billion, implies a revenue multiple of between 25x and 40x. For comparison, Nvidia — which generates actual profits at scale — trades at approximately 20x to 25x forward revenue in the same period. The market is being asked to assign a higher revenue multiple to a company that loses more than a dollar for every dollar it earns than it assigns to the most profitable AI infrastructure company in the world.

That premium is justified only if you believe that OpenAI will achieve the market dominance in AI products and infrastructure that would produce the gross margins required to support that valuation over time — and that it will do so before competitors with stronger balance sheets, different governance structures, and in some cases more advanced research capabilities displace it. Goldman Sachs, Morgan Stanley, and JPMorgan clearly believe the market will reach for that valuation, or they would not have agreed to lead the deal. Whether the public market agrees will depend on what the prospectus says, what the roadshow reveals, and what Sam Altman tells investors about the specific path from $14 billion in annual losses to profitability by 2029.

Conclusion

The OpenAI IPO filing is the most consequential single document in the AI industry's transition from a research conversation to a public markets event. It is also, for the moment, almost entirely unreadable by the public — because the filing is confidential, and the company has made clear that it will remain that way until roughly three weeks before investors are asked to price it.

What is already visible tells an interesting and uncomfortable story. The revenue growth is real and remarkable. The losses are equally real and equally remarkable. The governance structure is unlike anything that has previously appeared in a major technology IPO. The profitability timeline is three to four years away. The nonprofit that controls the board does not answer to shareholders. And the CEO whose judgment and vision drive the company holds no equity in it, which is either the most principled stance a technology founder has ever taken or the most unsettling detail in a prospectus full of unsettling details, depending on how you read it.

The S-1 will eventually explain all of this in the carefully calibrated language of SEC disclosure. Until then, the most honest summary of the OpenAI IPO is this: the company is asking the public to trust a trajectory it has not yet proved, governed by a structure it has never before submitted to public accountability, at a valuation that requires everything to go right. That is not necessarily a reason not to invest. It is a reason to read the prospectus very carefully when it finally arrives.