THE NUMBERS BEHIND THE VALUATION: $25 BILLION IN REVENUE, $14 BILLION IN LOSSES
The baseline figures reported around the time of the confidential filing describe a company in a state that has no clear historical precedent. Annualised revenue is approximately $25 billion — a figure that, for context, Salesforce required eighteen years to reach, and which OpenAI has reached in under a decade since its founding. The private market valuation of $852 billion, established in October 2025, was already the highest ever recorded for a private technology company. The target IPO valuation of over $1 trillion would make it the largest technology IPO by valuation since Alibaba's 2014 listing at $168 billion — a number that illustrates how different the scale of AI valuations has become from anything that preceded them.
The loss figure — approximately $14 billion annually against $25 billion in revenue — is where the prospectus will do its hardest work. A company losing money at this scale while generating revenue at this scale is not, in itself, unusual in the history of technology IPOs. Amazon was unprofitable for years. Uber burned through billions before reaching operating leverage. The standard argument for tolerating losses at this scale is that the company is investing at a pace that builds a durable competitive position — that the losses today purchase the margins of tomorrow. What makes OpenAI's case distinctive is the nature of the costs. Unlike a delivery platform whose losses reflect subsidised pricing to acquire customers, or a social network whose losses reflect engineering headcount and infrastructure, OpenAI's primary cost driver is compute. Training frontier models at scale, and serving inference at the volume OpenAI's products now require, consumes extraordinary quantities of GPU time. And unlike marketing spend or headcount, compute costs do not straightforwardly compress as revenue grows — they tend to grow alongside capability ambitions, and the company's stated mission requires it to keep pushing the capability frontier regardless of whether the unit economics on the current generation have been resolved.
The gross margin picture, which the S-1 will be required to disclose in detail, is the number that analysts will study most closely. If OpenAI's inference costs are declining faster than its revenue growth — meaning that each additional dollar of API and ChatGPT revenue is more profitable than the last — the loss story becomes a transition story, with an intelligible endpoint. If inference costs are sticky or growing alongside revenue, the loss story is structurally different, and the path to profitability requires either a step-change in model efficiency that has not yet materialised, or a revenue mix that tilts toward higher-margin products. The company has said publicly that it expects to reach operating profitability within two to three years. The S-1 will require it to back that claim with disclosure of the specific assumptions behind it.
WHAT GOLDMAN SACHS AND MORGAN STANLEY HAVE TO SELL
The choice of Goldman Sachs and Morgan Stanley as lead underwriters is a signal in itself. Both banks are among the largest facilitators of institutional equity distribution in the world, and their involvement suggests OpenAI intends to build a book with the largest sovereign wealth funds, mutual funds, and asset managers — not primarily with the retail investor base that has been the dominant buyer in some recent AI-adjacent IPOs. The roadshow pitch they will construct has several identifiable components, each of which carries a different degree of analytical credibility.
The growth story is the most straightforward. Revenue that has scaled from essentially zero to $25 billion in less than three years, with no obvious ceiling in sight, is a genuinely unusual asset. The customer base spans individual consumers paying $20 per month for ChatGPT Plus through to enterprises with multi-million-dollar API contracts. The API business has sticky economics: once engineering teams are building on GPT-5.5 or its successors, switching costs are real, and the network effects of developer tooling and ecosystem integrations accumulate over time. The enterprise contracts that have been disclosed publicly — Novo Nordisk's full drug pipeline on OpenAI, the financial services joint ventures, the healthcare deployments — represent a category of commitment that is difficult to unwind without significant operational disruption. These are not pilot contracts; they are production dependencies, which is precisely the customer relationship that generates the kind of predictable, durable revenue that equity markets value highly.
The competitive position story is more complicated and will require the most careful construction. Anthropic's revenue trajectory — described in this week's other disclosures as now potentially exceeding OpenAI's, with a lower training cost base — is a fact the S-1 cannot ignore. The prospectus will be required to name Anthropic, Google DeepMind, Meta AI, and xAI as competitive risks, and the risk factor disclosure on competitive dynamics will be among the most revealing sections of any S-1 in recent memory. How OpenAI characterises its own moat — whether it argues that the ChatGPT consumer brand provides a durable distribution advantage, or that GPT-5.5's capability position is not easily replicated, or that the Microsoft partnership creates an enterprise distribution lock-in — will tell the market something important about what the company actually believes, as opposed to what it says in press releases. The Microsoft relationship is itself complex: the exclusivity arrangement that ended earlier this month means OpenAI's models are now available on AWS and Google Cloud, expanding distribution but also removing one element of the competitive protection that the arrangement historically provided.
The governance story is the one that has no precedent in any prior technology IPO. OpenAI's corporate restructure — converting from a capped-profit limited liability company to a standard for-profit corporation — was a prerequisite for the IPO and was completed in early 2026 after a period of intense internal and external scrutiny. The non-profit entity that owns the original mission of "safe and beneficial AGI" retains a stake in the new for-profit, and the board of the non-profit retains certain oversight rights over the company's direction. The S-1 will be required to describe those rights in precise legal language, and the market will then have to decide what they are worth — both as a governance protection and as a potential constraint on the company's commercial latitude. This is genuinely novel territory. No major technology company has gone public with a governing document that explicitly subordinates certain commercial decisions to a mission defined in terms of the welfare of humanity, and the market's reaction to that structure — whether it is treated as a feature that signals responsible stewardship or a bug that introduces uncertainty into capital allocation decisions — is unknowable in advance.
WHAT THE IPO MEANS BEYOND OPENAI
The significance of the OpenAI IPO extends well beyond the company's own balance sheet and the immediate wealth creation event for its employees and investors. Public markets impose a specific and powerful kind of discipline: quarterly earnings calls, short-seller scrutiny, mandatory disclosure of material risks, and the continuous pressure of a stock price that reflects the aggregated judgement of every market participant about the company's prospects. Private AI companies have operated largely outside this discipline. The valuations assigned to Anthropic, xAI, and the earlier rounds of OpenAI itself have been set by investors who are sophisticated but who do not face the same accountability structures as public market shareholders. The OpenAI IPO will be the first major test of whether the valuations that private AI markets have established bear any relationship to what public markets, with full disclosure, are willing to pay.
The outcome of that test will cascade through the entire AI investment ecosystem in ways that are difficult to predict but easy to identify in advance. If the IPO prices at or above the $1 trillion target and trades up in the aftermarket, it will validate the private market valuations of every AI company in the current generation and likely accelerate investment into the next generation of labs. It will also generate enormous pressure on Anthropic to consider its own path to liquidity — the company's $950 billion valuation implies investor expectations that cannot be satisfied indefinitely by private secondary markets. If the IPO prices below the target, or struggles in aftermarket trading, the effect runs in the opposite direction: every AI company's private market valuation comes under immediate scrutiny, the cost of capital for the sector rises, and the timeline for the reckoning over whether frontier AI is a viable business at current investment levels is pulled forward.
The broader question the IPO forces is one that the AI industry has been able to defer while operating under private market conditions: is the frontier AI race, at its current pace and capital intensity, a rational allocation of resources? The argument for the current pace — made explicitly by figures like Sam Altman and implicitly by the capital commitments of SoftBank, Microsoft, and the sovereign wealth funds that have invested — is that the capability improvements being purchased by the current level of investment will generate economic value that justifies the cost by a large margin. The argument against is that the compute scaling thesis that drove investment through 2024 and 2025 has produced diminishing returns at the margin, that the economic value of capabilities above a certain threshold is speculative rather than demonstrated, and that the concentration of AI capability in a small number of heavily funded labs creates systemic risks that the market has not yet priced. The S-1 that Goldman Sachs and Morgan Stanley are preparing will not resolve this debate — no prospectus ever resolves a foundational question about a technology's economic trajectory. But it will force a level of public disclosure and analytical engagement with these questions that the private market phase of the AI industry has not required, and that engagement, whatever it produces, will be useful.