The headline number is $300 billion higher than Meta. Higher than Netflix, higher than Goldman Sachs, higher than most things humans have ever built. When SoftBank led a $40 billion funding round for OpenAI earlier this year, the implied valuation landed at $852 billion. Sam Altman didn't correct them.

But according to the Financial Times, something else was happening inside the company at the same time. Sarah Friar, OpenAI's CFO, was telling Altman they weren't ready to go public. Her concern: the gap between the narrative OpenAI tells investors and the actual financial reality was too wide. When your own CFO doesn't want to go public, that's worth paying attention to.

Friar didn't say this publicly. Nobody inside OpenAI does. But the Financial Times reported the internal disagreement, and it's the kind of friction that doesn't stay quiet for long.

The Numbers That Don't Add Up

OpenAI is projected to lose $14 billion in 2026. Not lose in the "burning cash to grow" startup sense. Lose because its costs are growing faster than its revenue, and its revenue is already enormous. The company is on track for roughly $12.7 billion in annualized revenue. The math still doesn't work.

The core problem is infrastructure. Running the models that power ChatGPT and the API costs extraordinary amounts of money. Microsoft has poured in billions in cloud compute. Even with that, the unit economics of serving AI at scale haven't resolved into something sustainable. Every time OpenAI lowers prices to compete, the losses get worse.

The product roadmap hasn't helped. OpenAI revised its plans at least twice in the past six months, TechCrunch reported. GPT-5 launched with less differentiation than expected. Operator features promised for enterprise clients shipped late. The company that once set every industry deadline is now scrambling to meet its own.

Then there's Anthropic. Dario Amodei's company went from $9 billion to $30 billion in annualized revenue run rate in roughly three months, according to Bloomberg. That's not a gradual climb. That's a company catching up fast. For the first time, Anthropic trades at a premium to OpenAI on secondary markets. Read that again: the company that was supposed to be the scrappy challenger is now valued more highly per dollar of revenue than the category leader.

Three Pressure Points

The IPO Window

Sam Altman wants to take OpenAI public in the fourth quarter of 2026. Sarah Friar thinks that's too soon. They are both right about something, which makes this harder to resolve.

Altman is right that the window won't stay open forever. Investor patience has limits. SoftBank, Microsoft, and others who put in billions at a $157 billion valuation in late 2023 and then again at $852 billion in 2025 need an exit path. Every month of delay increases the pressure on them and on Altman to justify the number.

Friar is right that $14 billion in projected losses is a difficult story to sell on a roadshow. Public market investors aren't venture capitalists. They want to see a path to profitability, not just a narrative about inevitable dominance. The current financials don't provide that path. An IPO with losses this large would require something close to a perfect pitch. And as Friar reportedly said, the narrative and the numbers aren't aligned.

The Enterprise Bet

OpenAI's strategy has shifted. ChatGPT made the company famous, but enterprise contracts are where it's trying to make money. Deals with corporations, government contracts, API arrangements with product companies: this is the new focus.

Enterprise has better margins than consumer. It also has a problem. Anthropic got here first. Claude has been the default enterprise choice for companies that need reliable, safe, consistent output. Anthropic built its brand specifically around enterprise trustworthiness. OpenAI is now competing on Anthropic's home turf, while simultaneously trying not to lose the consumer base that ChatGPT built.

The risk is obvious. Consumer momentum is hard to maintain when your attention is elsewhere. If ChatGPT user growth flattens while OpenAI chases B2B contracts, the company ends up with neither a dominant consumer product nor enterprise leadership. That's the squeeze Altman is trying to avoid.

The Talent Drain

Anthropic didn't build itself out of nothing. It was founded by Dario Amodei, Daniela Amodei, and roughly a dozen other people who left OpenAI. That departure happened in 2021. Since then, the talent flow hasn't reversed.

OpenAI has seen significant departures over the past year. Several senior researchers left for competitors or to start new companies. The compensation structure creates a specific vulnerability: much of the equity given to employees is tied to OpenAI's valuation. At $852 billion, that equity looks valuable. If the valuation corrects, the calculus changes. Employees who might have stayed for the upside start calculating their alternatives.

This is the compounding problem of an inflated valuation. It attracts capital on the way up. It creates retention risk on the way down.

What Happens Next

There are three plausible versions of how this plays out over the next 18 months.

The most likely outcome: OpenAI goes public in late 2026, but at a lower number. Something in the $500 billion to $600 billion range, after markets have digested the losses and priced in competition from Anthropic and Google. The company survives. It remains a major player. But the narrative shifts from "inevitable dominant AI company" to "one of several large AI companies." That's a meaningful change. The entire business model of recruiting talent, raising capital, and signing enterprise deals has relied on the first story.

The bull case: Enterprise revenue surprises to the upside in the next two quarters. GPT-6 launches with a clear, demonstrable advantage over Claude and Gemini. The IPO proceeds at or near the current valuation. OpenAI emerges as the clear winner with a defensible lead. This is what Altman is betting on. It requires execution that hasn't been consistently visible so far.

The bear case: The IPO gets delayed to 2027. Anthropic's revenue lead widens. A few high-profile enterprise clients switch to Claude. Key researchers leave. By the time the IPO happens, the valuation has already corrected 40 to 50 percent in secondary markets, and the public offering becomes a way to give existing investors a partial exit rather than a triumphant debut. OpenAI continues to exist and operate, but the story of its dominance has ended before the opening bell.

What To Watch

Five signals to track over the next six months:

  • Anthropic's next funding round valuation. If Anthropic raises at $80 billion or more, it confirms the secondary market premium is real and growing, not a temporary anomaly.
  • OpenAI S-1 filing timing. If a filing doesn't appear by September 2026, the Q4 IPO window is almost certainly gone. Watch for the registration statement, not press releases about "plans to go public."
  • ChatGPT monthly active user numbers. OpenAI stopped publishing these consistently. That's a data point. If the numbers were accelerating, they'd be publishing them.
  • Senior executive and researcher departures. Not mid-level churn. Watch for VP-level or above, or researchers with named papers who show up at competitors.
  • Enterprise contract announcements. Size and duration matter. A $5 million annual deal is different from a $200 million multi-year commitment. When OpenAI announces enterprise wins, look at the actual terms.

What Happens to the Story

OpenAI probably won't fail. The technology is real. The revenue is real. Sam Altman has navigated harder situations. A company with $12.7 billion in annualized revenue doesn't disappear just because it loses $14 billion in one year, especially with Microsoft and SoftBank in its corner.

The more precise point is this: $852 billion prices in a future where OpenAI wins clearly and completely. That future is no longer the obvious outcome. Anthropic is real competition. Google has resources that dwarf everyone in this industry. The enterprise pivot may work. It may not. GPT-6 may create distance. It may not. When the probability of winning clearly drops from very high to just possible, the valuation has to adjust. That's not a crisis. It's arithmetic.

My Opinion

The real risk to OpenAI isn't Anthropic. It isn't Google. It isn't even the $14 billion in losses. The real risk is that the company has become too big to be honest about what it doesn't know.

$852 billion means you can't afford uncertainty. Uncertainty makes investors nervous. Uncertainty makes enterprise clients pick a competitor. Uncertainty makes talent reconsider. So the company keeps projecting confidence: GPT-6 is coming, the enterprise pivot is working, the IPO is on track. Maybe all of that is true. But AI is nothing but uncertainty right now. Every benchmark that matters is contested. Every capability claim is debated. The best researchers in the field disagree on basic questions about where this is going. Pretending otherwise isn't strategy. It's liability.

Sarah Friar is telling Altman something true: the gap between the story and the numbers needs to close before they go public. The question is whether Altman believes closing the gap means better numbers, or just a better story. Those are different problems. They require different solutions. And right now, it's not clear which one he's working on.

Author: Yahor Kamarou (Mark) / www.humai.blog / 19 Apr 2026