OpenAI at $1 Trillion: How to Think About a Company You Can't Yet Investigate

Introduction

OpenAI is reportedly preparing for a public listing at a valuation north of $300 billion, with some projections placing it at $1 trillion by the time shares actually trade. If that happens, it would be one of the largest IPOs in history, and almost certainly the most consequential AI IPO to date.

The excitement is real. Generative AI is reshaping industries. ChatGPT reached 100 million users faster than any product in history. Revenue is growing at triple-digit percentages. Every headline suggests this is the defining technology company of the decade.

But excitement is not analysis. And when a company that burns $14 billion a year in compute costs asks you for a trillion-dollar valuation before you can read a single SEC filing, the right response is not to rush in. It is to slow down, identify the questions that matter, and prepare to investigate the moment the filings become available.

This article is not a recommendation to buy or avoid OpenAI. It is a framework for thinking clearly about a company whose economics are genuinely unprecedented, and whose corporate structure is unlike anything the public markets have seen before.

The Numbers Behind the Headline

Before exploring the questions investors should ask, it helps to understand the scale of what is being proposed.

Those four numbers, taken together, tell a story that should give any investor pause. Revenue is growing fast, but the company is spending more than it earns, the path to profitability is years away, and the market is being asked to price in a future that has not yet arrived.

This is not inherently disqualifying. Amazon was unprofitable for years. So was Tesla. But in both cases, investors could read 10-K filings, examine unit economics, track gross margins, and study the competitive landscape in detail. With OpenAI, as a private company, most of that information is unavailable to the public. What we have instead is a collection of reported figures from press coverage, investor presentations that have leaked, and statements from executives.

That is not the same thing as a 10-K. It is not audited. It is not subject to SEC comment letters. And it does not come with the risk factor disclosures that public companies are legally required to file.

A Corporate Structure Unlike Any Other

OpenAI began as a nonprofit research lab in 2015. In 2019, it created a “capped-profit” subsidiary to attract investment, with returns originally capped at 100x the initial investment. That cap has since been restructured.

As of early 2026, OpenAI is in the process of converting to a for-profit public benefit corporation. This conversion involves several layers of complexity that investors should understand:

• The nonprofit layer still exists. The original OpenAI nonprofit retains certain governance rights and will receive a significant equity stake in the new for-profit entity. How this equity is valued and how the nonprofit exercises its governance authority are open questions.
• The public benefit corporation structure. A PBC is legally required to balance shareholder returns with a stated public benefit mission. This is different from a standard C-corp. It means the board has a legal obligation to consider factors beyond pure profit maximization, which can affect capital allocation, pricing decisions, and competitive strategy.
• Historical investor agreements. Early investors had return caps and specific terms that are being renegotiated as part of the conversion. The terms of these renegotiations affect dilution and the economic rights of any new public shareholders.
• Microsoft's relationship. Microsoft has invested over $13 billion in OpenAI and has a complex revenue-sharing and licensing arrangement. The exact terms of this arrangement, including how Azure revenue from OpenAI models is shared, will be critical to understanding the true economics of the business.

None of this means the structure is bad. It means it is unusual. And unusual structures require more scrutiny, not less.

Revenue Is Not the Same as Economics

OpenAI's reported $12.7 billion in 2025 revenue is impressive by any standard. But revenue tells you what customers are paying. It does not tell you what it costs to serve them.

The Compute Cost Problem

Large language models are expensive to run. Every API call, every ChatGPT conversation, every enterprise deployment consumes GPU compute that OpenAI either owns or rents from cloud providers (primarily Microsoft Azure). The compute cost structure is the single most important economic variable in the business, and it is the one we know the least about.

Questions a 10-K filing would answer:

• What is the gross margin per API call at current pricing?
• How much does ChatGPT Plus ($20/month) actually cost to serve per user?
• What percentage of total revenue goes to compute costs versus other operating expenses?
• Are compute costs declining per unit of output as models become more efficient, or are they increasing as models get larger?
• What are the contractual terms with Microsoft for Azure compute? Are there volume discounts, minimum commitments, or take-or-pay provisions?

Without answers to these questions, revenue growth alone is not sufficient to evaluate the business. A company growing revenue at 200% year-over-year while losing money on every transaction is not the same as a company growing at 200% with expanding margins.

Enterprise vs. Consumer Revenue Mix

OpenAI generates revenue from multiple channels: ChatGPT Plus and Pro subscriptions, API access for developers, and enterprise contracts. These channels have very different economics:

• Consumer subscriptions are high-volume but potentially low-margin. Heavy users consume significant compute. Light users may churn.
• API revenue is usage-based and more predictable, but faces intense price competition from Anthropic, Google, Meta, and open-source models.
• Enterprise contracts typically come with higher margins and longer commitments, but require sales teams, custom deployments, and compliance infrastructure.

The mix matters enormously. A company where 70% of revenue comes from sticky enterprise contracts is valued very differently from one where 70% comes from consumer subscriptions that can be cancelled monthly.

The Circular Financing Problem

This is the issue that gets the least attention and arguably matters the most.

OpenAI's largest investor is Microsoft. Microsoft is also OpenAI's largest cloud computing provider (via Azure). And Microsoft is also one of OpenAI's largest customers and distribution partners (via Copilot, Bing, and Azure OpenAI Service).

Here is what that means in practice:

1. Microsoft invests billions in OpenAI.
2. OpenAI spends a large portion of that money on Microsoft Azure compute.
3. Microsoft recognizes that spending as Azure revenue.
4. Microsoft uses OpenAI models in its own products, generating revenue that partially flows back to OpenAI via licensing.
5. That revenue enables OpenAI to raise at a higher valuation.
6. Which enables Microsoft to mark up its investment.

This is not fraud. It is not illegal. But it is a financing structure where money flows in a circle, and each participant in the circle books revenue or gains from the flow. Traditional valuation methods struggle with this because they assume revenue represents independent demand from arms-length customers.

The question is not whether the Microsoft relationship is valuable. It clearly is. The question is: if you remove the circular flows, what does the underlying business look like? What is the organic demand from customers who are not also investors, cloud providers, and distribution partners?

The Competition Nobody Prices In

The trillion-dollar valuation thesis assumes OpenAI maintains a dominant position in AI. But the competitive landscape has shifted dramatically in the past 18 months.

Well-Funded Direct Competitors

• Anthropic (Claude) has raised over $10 billion and is backed by Amazon and Google. Its models consistently benchmark at or above GPT-4 class performance.
• Google DeepMind has effectively unlimited compute via Google Cloud, decades of AI research, and distribution through Search, Android, and Workspace.
• Meta AI releases open-source models (Llama) that enable anyone to run competitive inference at a fraction of the cost.
• xAI (Grok) has raised billions and has exclusive access to X (Twitter) data for training.
• Mistral, Cohere, and dozens of others compete in the API and enterprise market with differentiated offerings.

The Open-Source Threat

Meta's Llama models and the broader open-source ecosystem represent an existential competitive threat to any company whose business model depends on charging for model access. When a comparable model is available for free, pricing power erodes.

This does not mean OpenAI loses. It means the competitive moat is narrower than the valuation implies. A company valued at $1 trillion needs not just a good product, but a durable competitive advantage that justifies that multiple for decades. Whether OpenAI has that is an open question.

Customer Switching Costs

In traditional enterprise software, switching costs are high. Migrating from Salesforce or Oracle is painful and expensive. In AI, switching costs are meaningfully lower. An application built on the OpenAI API can be migrated to Anthropic, Google, or a self-hosted model with relatively modest engineering effort. The model is increasingly a commodity; the application layer is where differentiation lives.

This matters because investor assumptions about retention rates and pricing power should account for how easy it is for customers to leave.

Eight Questions to Ask Before You Buy

If and when OpenAI files an S-1 and you are considering buying shares, these are the questions the filing should help you answer. If the filing does not adequately address them, that itself is informative.

What Due Diligence Looks Like on Day One

When OpenAI does file its S-1 and begin trading, the investigative process starts immediately. Here is what a thorough stock due diligence review would cover:

SEC Filing Analysis

• Read the S-1 cover to cover. Not the summary. The whole document.
• Focus on the risk factors section. Companies are legally required to disclose every material risk they know about. This is where you find what the headlines omit.
• Study the related-party transactions section for the Microsoft relationship details.
• Examine the financial statements and footnotes, particularly revenue recognition policies, stock-based compensation, and off-balance-sheet obligations.
• Cross-reference management's discussion and analysis (MD&A) with the actual numbers. When management's narrative and the financial data tell different stories, trust the data.

Insider Activity

• Track Forms 3, 4, and 5 on SEC EDGAR from the day they begin filing.
• Watch for insider selling in the first six months after lockup expiration. If the people who know the company best are selling, ask why.
• Pay attention to the ratio of buys to sells among senior management.

Competitive Landscape

• Compare pricing, performance benchmarks, and customer acquisition costs against Anthropic, Google, and open-source alternatives.
• Track API pricing trends across all major providers. Declining prices across the industry signal commoditization.
• Monitor patent filings at USPTO for all major AI competitors to understand where R&D investment is focused.

Court Records and Regulatory Actions

• Search PACER for any pending litigation. Copyright lawsuits from content publishers, data licensing disputes, and employment cases are already in the system for several AI companies.
• Monitor regulatory proceedings in the EU, UK, and US. Enforcement actions and consent decrees can materially affect operations.
• Track the New York Times, Authors Guild, and other copyright cases that could set precedent for training data liability.

The Takeaway

OpenAI may turn out to be worth $1 trillion. It may turn out to be worth more. Or it may turn out to be a company that changed the world but never generated the returns that the IPO price implied. History has examples of all three outcomes.

The point is not to predict which outcome occurs. The point is to recognize that right now, before the S-1 is filed, we do not have the information required to know. We have revenue headlines, valuation estimates, and a lot of enthusiasm. What we do not have is:

• Audited financial statements with granular cost breakdowns
• Legal risk disclosures written under SEC penalty of fraud
• Related-party transaction details showing the true Microsoft economics
• Insider compensation and equity dilution data
• Off-balance-sheet obligations and contractual commitments
• A complete risk factor disclosure

All of that arrives with the S-1. Until then, the responsible approach is to prepare your questions, understand the structure, and resist the pressure to form a conviction based on incomplete data.

When the filings land, investigate. Not before.