What an S1 Actually Is and Why It Matters
Before a company can sell shares to the general public, it must file an S1 registration statement with the Securities and Exchange Commission. The document is a required disclosure , revenue figures, operating losses, risk factors, governance structure, material contracts, and anything else a reasonable investor would need to make an informed decision. It is written by lawyers, reviewed by regulators, and legally binding in ways that press releases and investor presentations are not.
What a company chooses to include in its S1 tells you what management wants investors to focus on and believe. What a company describes in vague or broad terms tells you what it would prefer investors not examine too closely. Both types of information are useful. OpenAI's filing contains meaningful amounts of both.
This document is public. Anyone can read it. The numbers are real in a way that no prior disclosure about OpenAI's financials has been , the company was private and not required to report publicly until this filing. What it reveals changes the quality of information available about one of the most consequential companies of the current moment.
The Revenue and the Loss, Held Together
OpenAI reported $3.7 billion in revenue for 2024. That is a large number growing at a rate that would make almost any investor pay serious attention. The company is selling a meaningful quantity of something a very large number of people and organisations want badly enough to pay for. The revenue growth trajectory is real.
The same year, OpenAI reported operating losses of approximately $5 billion. Revenue of $3.7 billion and losses of $5 billion means the company spent roughly $8.7 billion to generate $3.7 billion in revenue. The gap between what comes in and what goes out is not small, and it is not narrowing proportionally to revenue growth , at current spending rates, the losses are growing roughly in proportion to or faster than the revenue.
This is not automatically a sign of a failing business. Early Amazon operated at losses for years while building infrastructure and market position that eventually generated substantial margins. Early Uber spent far more than it earned for years while building network density. The question is always the same: is the spending building something with durable structural advantages that will generate better margins once the growth phase matures? The S1 offers a theory of how that happens. It does not prove the theory is correct , it cannot, because the outcome is in the future.
The honest position is that OpenAI is a very fast-growing business that is also spending at a rate that would exhaust most organisations within a few years without continued external capital. The IPO is partly about building a revenue and profit business , and partly about accessing public market capital to fund the spending while that profit business develops.
The Path to Profitability the S1 Describes
OpenAI's S1 identifies three specific levers for closing the gap between revenue and spending. First: inference cost reduction. The cost of running a model , the compute required to generate each response , has been falling as hardware improves and engineering optimisation advances. If this decline continues, the margin on each conversation or API call improves without needing to raise prices. The unit economics get better with scale and time.
Second: enterprise customer expansion. Enterprise contracts carry substantially higher margins than consumer subscriptions in most software businesses, and this appears true for AI products as well. A business paying for API access or a custom enterprise deployment with volume commitments generates more margin per dollar of revenue than an individual paying $20 per month for ChatGPT Plus. The shift in revenue mix toward enterprise improves the overall margin profile even without changing the cost structure.
Third: model efficiency improvement. Smaller models that can perform the same quality tasks as larger ones cost less to serve at inference time. The trend toward capable small models , models that perform tasks previously requiring much larger parameter counts , reduces the infrastructure cost per unit of useful work. If this trend continues, the biggest cost driver in the business gets cheaper as the models get better.
All three of these levers are plausible. The trends supporting each of them are real and documented. None of them is guaranteed to move fast enough, or in the right combination, to close the gap within a timeframe that makes the current valuation defensible. The S1 presents them as a credible path because they are credible , and because that is the job of the document.
What the Valuation Math Actually Implies
OpenAI is seeking a valuation of approximately $300 billion at IPO. At $3.7 billion in revenue, that is an 81x revenue multiple. Comparable software-as-a-service companies growing at high rates , fast-growing enterprise software with strong retention , typically trade at 5 to 15 times revenue in public markets. The premium OpenAI is seeking is large enough that it cannot be explained by growth rate alone and requires a different kind of argument.
The argument offered is the "platform" thesis: OpenAI is not primarily a software company that makes chatbots competing in a market that will commoditise. It is the infrastructure layer for the AI economy , the way Amazon Web Services became the infrastructure layer for cloud computing, collecting margin from the full ecosystem of applications built on top of it. If that thesis is right, the addressable market is measured in trillions and the current revenue significantly understates the eventual scale.
If the thesis is wrong , if OpenAI turns out to be a software company with a high-quality product in a market where multiple well-funded competitors are building similar products, and where model costs continue declining toward commodity pricing , the 81x multiple looks very difficult to sustain. The valuation is a bet on which of those two futures arrives and on what timeline. The S1 does not resolve this question. It gives investors the numbers to form their own view of it.
The Risk Disclosures That Are Genuinely Unusual
S1 risk factors are carefully written for legal reasons , companies list risks they consider material because those disclosures provide some protection if things go wrong and investors later claim they were not warned. The standard is what a reasonable investor would consider important to know. Companies have strong incentives to include real risks rather than fabricate them, but also to describe those risks in ways that are not unnecessarily alarming.
OpenAI lists "model misalignment" and "catastrophic AI risk" as business risks. This is the first major technology IPO to include AI safety concerns as formal material risk factors in an S1 filing. The document is not saying these outcomes are likely. It is saying they are real enough possibilities that a reasonable investor should factor them into an investment decision.
The governance disclosures are also unusual in ways worth examining carefully. The for-profit conversion created a structure where the original non-profit entity retains equity in the new for-profit company and holds certain governance rights over it. This is a novel arrangement without clear legal or corporate precedent. The S1 describes it in terms that satisfy regulatory disclosure requirements without fully resolving how conflicts between the non-profit's mission obligations and the for-profit's shareholder obligations would be handled in practice.
For a company seeking a $300 billion valuation, governance clarity is a material concern. The S1 discloses the structure. It does not fully answer the question of how it functions under stress.
What the Document Omits
The Microsoft partnership terms are described in broad strokes beyond what was already public knowledge. The revenue-sharing arrangements between OpenAI and Microsoft , which are material to understanding how much of OpenAI's revenue it actually keeps, and under what conditions , are characterised in general terms. This is likely a combination of commercial confidentiality and legal agreement, but it means investors cannot fully model the business without knowing those terms.
The competitive dynamics section describes the market landscape without quantifying the threat from Google, Anthropic, Meta's open-source models, or the growing number of well-funded competitors. This is normal for an S1 , companies do not typically quantify how dangerous their competitors are in their own IPO filings , but it means the competitive picture investors are reading is the one OpenAI chose to paint.
What the S1 does well is give investors accurate numbers to work with and an honest framing of the central uncertainty. OpenAI is spending more than it earns. The path to changing that depends on trends that are plausible but not guaranteed. The company believes it is building something platform-level. The market will decide whether to price that belief at $300 billion.
The numbers in the document are real.
The story attached to them is a bet about what the company becomes.
The S1 makes that bet explicit , which is exactly what it is required to do, and more transparency than most investors had before it was filed.