The Vibes-Based Valuation

In March 2026, three frontier AI models launched in a single month: GPT-5.4, Gemini 3.1 Ultra, and Grok 4.20. The capability gap between labs compressed from months to weeks. On most benchmarks — the ones we've shown are compromised, contaminated, and self-graded — the models are nearly indistinguishable. Yet OpenAI is valued at $500 billion. Anthropic at $350 billion. Meta's Llama — open-source, free, competitive on benchmarks — at zero.¹

Nothing in the current financials explains a $500 billion valuation.

The Multiple

OpenAI's $500 billion valuation looked like 167x revenue when early 2025 projections pegged revenue at $3 billion. Revenue grew faster than expected — CFO Sarah Friar confirmed $20 billion+ annualized by late 2025 — putting the real multiple closer to 25-38x. That's still 3-7x higher than even the most aggressive AI startup multiples (40-50x for generative AI), and far above traditional SaaS (5-10x).²

Even at 25x, the multiple assumes revenue growth that hasn't been sustained quarter over quarter. And the cash burn that funds that growth is accelerating faster than the revenue.

Anthropic's valuation is lower in absolute terms but arguably more defensible. At $350 billion, with annualized revenue hitting $19 billion in March 2026 (up from a run rate topping $9 billion in January 2026), Anthropic is growing at roughly 10x per year — nearly three times OpenAI's 3.4x growth rate.³ The market is pricing Anthropic's trajectory and OpenAI's position. Whether either price is justified depends entirely on whether you believe the current growth rates are sustainable.

The Cash Burn

OpenAI's annual cash burn is projected to rise from $17 billion in 2026 to $35 billion in 2027, peaking at $47 billion in 2028. The company needs between $50 billion and $80 billion in fresh capital to sustain operations through profitability.⁴

Read that again. The most valuable AI company in the world needs another $50-80 billion just to survive long enough to become profitable. Its marquee consumer product, Sora, burned $15 million per day before being shut down. Its path to profitability assumes revenue growth that, at current rates, won't close the gap before the capital runs out.

Anthropic's burn is smaller — an estimated $5.2 billion EBITDA loss in 2025 — but the company just raised $30 billion in a single round to fund it.³ Both companies are spending more than they make. Both are betting that future revenue will justify present losses. The historical precedent for this bet (Amazon, Uber, Tesla) is that it sometimes works. The historical precedent for the current AI multiple environment — 25-38x revenue — is that it usually doesn't.

The Convergence Problem

Three frontier models in one month. The capability gap compressing to weeks. This is the opposite of what justifies premium valuations.

In most technology markets, convergence drives commoditization. When every product does roughly the same thing, pricing power evaporates. The AI industry is watching this happen in real time — and the valuations haven't adjusted.

As we documented in The Benchmark Industrial Complex, the benchmarks used to differentiate models are compromised. Companies cherry-pick evaluations, contaminate test data, and report only the scores where they lead. When every model claims "state of the art" on a curated selection of tests, the term becomes meaningless. And when the term that justifies the valuation becomes meaningless, what's left?

Distribution, brand, enterprise contracts, and lock-in. The things that have always determined technology market share once the technology itself commoditizes. OpenAI has ChatGPT's consumer mindshare. Anthropic has the enterprise trust and the safety brand. Meta has open-source adoption. Google has search integration. None of these are worth 25-38x revenue. But they're worth something — and the question is what.

The Sora Signal

Sora's shutdown is the clearest signal that the economics haven't caught up to the valuations. OpenAI killed its most-hyped consumer product because it couldn't make the math work — $15 million per day in costs, $2.1 million lifetime revenue.⁵

If OpenAI can't make video generation economically viable, how much of the rest of the product portfolio is similarly subsidized? How much of the $30 billion revenue target depends on features that cost more to serve than they generate? The company doesn't break out unit economics by product line. The investors don't seem to be asking.

PitchBook's assessment: "The market has rewarded story over substance. Companies carrying the highest valuations are built on the weakest business fundamentals."²

What's Being Priced

The honest answer is: a future that may or may not arrive.

The $500 billion valuation assumes OpenAI achieves and sustains market dominance in a market that's rapidly commoditizing, reaches profitability before burning through $50-80 billion more in capital, maintains pricing power as open-source alternatives close the capability gap, and converts a consumer brand into enterprise revenue at margins that justify the multiple.

Each of these assumptions is plausible. All of them together, at the required scale, in the required timeframe, is a bet. The market is pricing a bet.

As we wrote in The AI Washing Index: saying "AI" on an earnings call adds 8.2 percentage points to your stock price. Being an AI company at this moment in history adds multiples that no business metric can support. The question isn't whether AI will be transformative. It probably will. The question is whether the specific companies currently valued at hundreds of billions will be the ones that capture the value — or whether, like Pets.com and Webvan, they're the right idea at the wrong price.

25-38x revenue. $17 billion annual burn. Sora dead. Benchmarks gamed. Models converging. The vibes are excellent. The math is not.