The $852 billion question

Eight hundred and fifty-two billion dollars. That is the number OpenAI put on itself when it closed its latest funding round on March 31, 2026. To put it in perspective, Boeing is worth about $130 billion. Goldman Sachs hovers around $180 billion. Disney, Nike, Starbucks, all smaller. OpenAI, a company that has never turned a profit and projects a $14 billion loss this year, is now valued higher than most of the Fortune 100. The round raised $122 billion in committed capital, the largest private fundraise in Silicon Valley history, co-led by SoftBank alongside Andreessen Horowitz, D.E. Shaw Ventures, MGX, TPG, and T. Rowe Price. Amazon, Nvidia, and Microsoft all participated. For the first time, OpenAI opened participation to individual investors through bank channels, pulling in another $3 billion from retail money. The sheer scale of these numbers demands a question that nobody backing this deal seems interested in answering honestly: what exactly are investors buying?

The case for $852 billion

The bull case is not baseless. OpenAI now generates $2 billion per month in revenue. Annual revenue hit $13.1 billion in 2025, and the annualized run rate recently crossed $25 billion. ChatGPT has over 800 million weekly active users. More than 9 million paying business users rely on it daily. Codex, OpenAI's AI coding tool, has tripled its weekly user base since the start of 2026 to 1.6 million. By any normal measure, this is extraordinary growth. OpenAI is also telling investors it expects $280 billion in total revenue by 2030, split roughly evenly between consumer and enterprise. If you believe that projection, the current valuation starts to look less absurd. You are not paying for what OpenAI earns today. You are paying for a future where AI becomes the operating layer of the global economy, and OpenAI is the company that controls it. The problem is that "if" is doing an enormous amount of work in that sentence.

What OpenAI actually has

Strip away the narrative and look at what is concretely in the building. First, the models. OpenAI's GPT series remains among the best in the world, but the moat around model quality has been eroding for over a year. Competitors ship comparable capabilities within weeks of each release. Google's Gemini 3 matches or exceeds GPT on most benchmarks. Anthropic's Claude has become the default in enterprise and coding. Open-source models from Meta, Mistral, and others continue to close the gap. Model quality alone is not a durable advantage when the frontier moves this fast. Second, the API business. Revenue is growing, but margins are thin. Running inference at scale is extraordinarily expensive, and OpenAI is locked in a price war with every other provider. API pricing has dropped by orders of magnitude over the past two years, and competitors are undercutting on cost. The API is becoming a commodity business, which is not what you want when your valuation implies platform-level margins. Third, ChatGPT. The consumer product is OpenAI's strongest asset, with genuine mindshare and habit formation. But it faces mounting pressure. Google is embedding Gemini into Search, Gmail, Docs, Chrome, and the entire Workspace suite, reaching 750 million monthly AI users without asking anyone to download a new app. Anthropic is winning developer loyalty with Claude Code. Apple is integrating AI directly into its hardware layer. ChatGPT is a standalone product competing against AI that is woven into products billions of people already use. JPMorgan described OpenAI's competitive moat as "increasingly fragile." OpenAI's own internal strategy memos reportedly acknowledge that platform giants could block ChatGPT or push their own assistants without giving users fair alternatives. That is not the language of a company sitting comfortably on an $852 billion throne.

The vision premium

So if the models are not a moat, the API is a commodity, and the consumer product is under siege, what justifies the valuation? The answer is a story. Specifically, the story that OpenAI will become the platform, that it will be to intelligence what Google is to search or what Microsoft is to enterprise software. This is not a new playbook. In 2019, WeWork was valued at $47 billion on the premise that it was not a real estate company but a technology platform that would redefine how the world works. Investors were not buying office leases. They were buying a vision. When the IPO process forced actual scrutiny, the gap between narrative and reality collapsed the valuation to under $10 billion within months. By 2023, WeWork was bankrupt. OpenAI is not WeWork. The technology is real, the users are real, and the revenue is real. But the valuation structure rhymes. At roughly 34 times its annualized revenue, OpenAI trades at a multiple that assumes not just continued dominance but an expansion of that dominance into territory it does not yet occupy. Investors are buying a story about the future, not a spreadsheet about the present. The difference between a vision premium and a bubble is whether the vision materializes before the money runs out. OpenAI projects cumulative losses of $218 billion between 2026 and 2029. It has committed to roughly $600 billion in compute infrastructure spending by 2030. These are numbers that make the WeWork cash burn look quaint.

At least Nvidia sells shovels

Contrast OpenAI's position with Nvidia's. The AI capex boom is projected to exceed $700 billion, and Nvidia sits at the center of it. Every dollar that OpenAI, Google, Microsoft, and Anthropic spend on AI infrastructure flows disproportionately through Nvidia's supply chain. Nvidia currently consumes roughly two-thirds of the world's advanced chip packaging capacity. It is, quite literally, selling shovels during a gold rush. The circular dynamics are hard to ignore. Nvidia invested up to $30 billion in OpenAI. OpenAI uses that capital to buy Nvidia GPUs. Nvidia reports record revenue, its stock rises, and it can invest even more. Meanwhile, OpenAI points to its growing compute capacity as evidence of demand, which justifies its next fundraising round. Analysts have compared this to the vendor financing that fueled the dot-com bubble, where telecom equipment makers lent money to their own customers to buy their products. But even within this loop, Nvidia's position is fundamentally different. Nvidia has actual profit margins. It sells a physical product that every participant in the AI race needs. If the AI buildout slows, Nvidia's growth slows, but the company still has a real business underneath. OpenAI's position is more precarious. It is a customer in this ecosystem, not a supplier. It is spending the money, not collecting it.

Distribution or mindshare?

This is the question I keep coming back to. OpenAI has mindshare. Everyone knows ChatGPT. It is the default name people use when they talk about AI, the way "Google" became a verb for search. But mindshare is not the same as distribution. Distribution means being embedded in the workflows people already use, in ways that are hard to rip out. Google has distribution: Gemini shows up inside Gmail, Docs, and Search without users making a conscious choice. Microsoft has distribution: Copilot is built into Office, Teams, and Windows. Apple has distribution: AI runs on the device in your pocket. OpenAI has a chat window. A very popular chat window, but a chat window nonetheless. The company is trying to build distribution through enterprise sales, through partnerships, through its Frontier platform for enterprise AI deployment. But building a platform from scratch while burning $14 billion a year is a fundamentally different challenge than bolting AI onto an empire that already exists. The secondary market is already reflecting this tension. Bloomberg reported that in the weeks surrounding the $852 billion round, about half a dozen institutional investors approached secondary marketplaces looking to sell roughly $600 million in OpenAI shares. Demand for OpenAI stock on the secondary market has been dropping, even as Anthropic's shares run hot. Smart money is not unanimously convinced.

The commodity question

Here is the uncomfortable truth that the $852 billion valuation has to confront: intelligence is becoming a commodity. Every few months, the gap between the best model and the second-best model narrows. Pricing drops. Capabilities converge. A year ago, GPT-4 had a clear lead. Today, there are half a dozen models within striking distance on any benchmark you care about, and open-source alternatives are closing in fast. If intelligence itself is a commodity, meaning cheap, abundant, and interchangeable, then what justifies $852 billion for one supplier? You do not pay 34 times revenue for a commodity business. You pay that multiple for a platform with lock-in, network effects, and pricing power. OpenAI's challenge is that it has not yet built any of those things in a way that competitors cannot replicate. The bet could still pay off. If OpenAI cracks autonomous agents at scale, if it builds the platform that enterprises cannot live without, if it achieves the kind of integration and lock-in that justifies a multi-hundred-billion-dollar valuation, then the current price will look like a bargain. That is a real possibility, and dismissing it would be intellectually dishonest. But it is also possible that we are watching the most expensive vision premium in the history of technology. $852 billion for a company that has never been profitable, that faces existential competition from three of the largest corporations on Earth, and that is selling a product whose core capability gets cheaper and more widely available by the quarter. The investors writing these checks are not stupid. But they are making a bet on a specific future, and the gap between that future and the present is wider than $852 billion can comfortably bridge. The question is not whether OpenAI is a good company. It is. The question is whether any single AI company is worth $852 billion in a world where intelligence is rapidly becoming the cheapest resource in the stack. The market seems very sure of the answer. I am less convinced.

References