OpenAI's $1 Trillion IPO: Wall Street's Biggest Bet on AI's Unproven Economics

Sometime before the end of summer, investors will be asked to decide whether OpenAI is worth more than $1 trillion. The company filed a confidential S-1 with the Securities and Exchange Commission in late May, initiating the regulatory process for what would be the largest technology IPO in history. Goldman Sachs and Morgan Stanley are leading the deal, with Citigroup and JPMorgan Chase joining the syndicate — a roster that signals just how seriously Wall Street is treating the event.

The target valuation — between $852 billion and $1 trillion based on current discussions — would put OpenAI above Saudi Aramco’s 2019 IPO ($1.7 trillion was the headline, but that was a Saudi domestic listing engineered for political purposes) in terms of genuine public market ambition from a technology company. For context, when Google went public in 2004, it was worth $23 billion. OpenAI wants to list at 43 times that.

The Revenue Story: Real and Spectacular

The bull case for OpenAI begins with a revenue trajectory that has few parallels. The company now generates approximately $2 billion per month — $24 billion on an annualized basis — up from $1.6 billion monthly at the start of the year. ChatGPT serves over 1 billion weekly active users (a milestone the company announced last week), with enterprise contracts — OpenAI’s Teams and Enterprise plans — now driving more than 40% of total revenue.

The product lineup is expanding rapidly. ChatGPT’s recent launch of a self-serve advertising platform creates a new revenue stream that could compound significantly as user engagement continues to grow. Codex, OpenAI’s coding agent deployed inside developer platforms and on AWS Bedrock, is generating software automation revenue at enterprise scale. The voice API — now in its second generation — is powering customer-facing conversational AI applications across industries from healthcare to financial services.

API revenue from developers and enterprises represents the segment that most resembles a traditional SaaS business: recurring, sticky, and growing. The company’s O3 reasoning API, used by researchers, lawyers, and financial analysts for complex multi-step inference, commands premium pricing that has created a high-margin tier within an otherwise cost-intensive business.

The Cost Problem: Losing $1.22 for Every Dollar Earned

The bear case is arithmetic. In 2025, OpenAI burned approximately $22 billion to earn $13 billion, producing losses that would stun even the most loss-tolerant venture investor in any other decade. The company is projecting operating losses of around $14 billion for 2026 — an improvement over 2025 but still a figure that, if sustained, would consume even a $1 trillion market cap at a disturbing pace.

The fundamental economics of frontier AI training are brutal. Each new model generation requires compute investments in the billions. Inference costs — the expense of running the model in response to each query — scale with usage, which means OpenAI’s best business outcome (more users, more queries) also produces more losses absent a dramatic improvement in efficiency or a step-function reduction in hardware costs.

CEO Sam Altman has argued consistently that the cost curve will bend downward as model efficiency improves, custom silicon replaces rented data center capacity, and the company’s infrastructure investments achieve scale. OpenAI’s partnership with Microsoft — which retains a 49% stake in the company’s commercial operations through 2032 — gives it preferential access to Azure infrastructure, reducing the effective compute cost below what a standalone company would pay.

But the S-1 will need to explain, in terms that public market investors will accept, a credible path to profitability that goes beyond “the technology will get cheaper.” Fund managers who can buy Nvidia — which is profitable, growing, and central to the AI stack — will need a compelling reason to own a company that is losing money at scale.

The Structural Complexity: Not a Normal Company

OpenAI’s corporate structure adds another layer of complexity to the IPO narrative. The company is in the final stages of a restructuring that converts it from a capped-profit LLC (a structure that limited investor returns) into a public benefit corporation — a standard Delaware structure more compatible with public markets.

The capped-profit structure had assigned different return rights to different investors, with Microsoft’s stake structured differently from venture investors like Thrive Capital. The restructuring, which the company completed earlier this year, normalizes these relationships but required months of negotiations and some renegotiation of terms. The S-1 will need to clearly explain what shareholders are actually buying, and what rights they receive relative to the original nonprofit parent entity that Sam Altman and the board retain control over.

That nonprofit parent — the original OpenAI Inc. — holds a perpetual license to the commercial company’s technology and board seats with oversight authority. Exactly how much operational independence a public company has when a nonprofit parent with divergent objectives sits above it will be one of the questions analysts will probe hardest.

What the IPO Race Means for the Industry

OpenAI’s filing comes just weeks after Anthropic filed its own confidential S-1, setting up a potential race between the two most valuable AI companies to reach public markets. Both are targeting September 2026. Both are valued above $800 billion. Both are losing money.

The juxtaposition is instructive about the state of the industry. The two most well-resourced, most technically capable frontier AI labs are simultaneously preparing the largest technology IPOs in history while running operating losses that would sink conventional technology companies. They are doing this because the strategic logic of AI development — capture scale, build data advantages, establish API dominance — justifies the spending, at least in a world where capital is available and the technology is advancing.

Whether public markets will sustain that logic is a different question. Retail investors have never been asked to price in the economics of training frontier models. The roadshow presentations will need to articulate a future state — not just a present trajectory — in which the company’s revenue exceeds its spending by a margin sufficient to justify equity ownership at trillion-dollar scale.

The Analyst View

Analysts at Goldman Sachs reportedly modeled the offering at a range of outcomes, with a base case valuation around $900 billion contingent on continued revenue growth above 30% year-over-year and evidence of improving gross margins on API revenue. The bull case — $1.2 to $1.5 trillion — depends on a faster-than-expected improvement in inference efficiency and continued enterprise expansion.

The bear case, at $500 to $600 billion, assumes public market investors apply a significant discount to the company’s losses relative to private market norms, and that regulatory risk materializes in ways that constrain product expansion.

September is a long time from now in AI. OpenAI will ship at least one major model before its roadshow begins. The company that investors see on the road may look different from the one that filed the S-1 in May — almost certainly in ways that make the story better, if the past two years of OpenAI’s trajectory are any guide.

The trillion-dollar question is whether that’s enough.