Anthropic Delays $2 Trillion IPO as AI Market Sends Warning Signals

By Nathan Brooks

Anthropic has pushed its IPO from October to November, and the delay says more than any roadshow slide. The company is still targeting a $2 trillion valuation — 435 times its 2025 revenue of $4.6 billion — but the market around it is flashing red on at least three fronts.

The numbers behind the offering are extraordinary. Anthropic's Q3 2026 revenue hit a reported $16.8 billion, up 45% quarter on quarter. That puts the annualized run rate at roughly $67 billion. Yet the leaked prospectus, reviewed by Reuters, also shows the company spent $7.33 billion on compute and infrastructure in 2025 — more than its total revenue — and posted a $42 billion net loss, of which $34 billion was a noncash accounting charge tied to financing instruments that could convert into shares.

What does the S-1 actually commit Anthropic to?

The heaviest line item is $518 billion. That's what Anthropic expects to spend over a decade on AI infrastructure with six partners including Google, Microsoft and Amazon. About 80% of that total is noncancelable or payable regardless of usage, according to Reuters' review of the confidential document.

One more buried figure: a quarter of Anthropic's 2025 revenue came from just two unnamed customers.

What signals is the market sending?

The delay isn't happening in a vacuum. Three distinct warning lights went off in recent weeks.

• NYC Comptroller Mark Levine sent a letter to CEO Dario Amodei urging Anthropic to pause the IPO entirely. The city's pension funds reportedly hold roughly $7 million in Anthropic exposure today — but could hold dramatically more if the stock enters the Russell 3000. Levine cited the same risks Anthropic's own prospectus flags: autonomous AI systems that "pose existential risks to humanity."
• Sam Altman delayed OpenAI's IPO. OpenAI will not go public in 2026. Altman called an IPO at this moment "ill-advised" given AI safety concerns — partly strategic positioning, but a real signal when the CEO of the second-most valuable AI company decides to wait.
• A $2 trillion chip-stock selloff last week was driven partly by concern over AI-buildout costs. Gartner's forecast explains the worry: through 2030, it sees at most $3.6 trillion in revenue attributable to AI data centers, versus $8.5 trillion needed to cover planned investment, financing costs and a 12% return.

Can closed models defend their pricing?

Here's the structural problem no quarterly growth figure resolves: Anthropic's $2 trillion valuation requires closed-model AI to keep a durable pricing premium over open-weight alternatives. That premium is eroding fast.

Enterprise adoption of open-weight AI rose from 23% to 67% in 2026, with companies reporting 70–90% cost savings. DeepSeek R1 delivers frontier-level performance on most enterprise benchmarks at 7 cents per million input tokens; independent analysis puts Anthropic's comparable input and output pricing at roughly 18x and 23x higher. DeepSeek's weights are free to download, letting enterprises self-host with no API contract — and the capability gap with closed models reportedly closes within six months of a frontier release.

Research firm New Constructs, which called the offering "an unprecedented test of investor gullibility," calculated that justifying $2 trillion would require an immediate 10% after-tax operating margin sustained for a decade and $358 billion in annual after-tax operating profit by 2035.

Why 2026 isn't 1999

The dot-com bubble had a genuine excuse — nobody knew what the internet would become. This wave doesn't. Gartner forecasts worldwide AI spending of $2.59 trillion in 2026, up 47% year over year. Yet a July 2025 MIT NANDA report estimated 95% of organizations it examined were getting no return from generative AI, and KPMG found just 7% of surveyed leaders say they've established ROI from AI.

So investors are being asked to pay 435x 2025 revenue for a company with $518 billion in locked infrastructure commitments, 25% revenue concentration in two unnamed clients, and a pricing moat that open-weight models close in six-month cycles.

What does the delay actually change?

Not much, in hard numbers. It adds one quarter of revenue data. It doesn't touch the commitments, the concentration or the margin math. As analyst Anurag Gurtu argues, the winners of this correction won't necessarily own the most expensive model — they'll own the architecture that turns any model into governed, auditable outcomes. The dot-com crash didn't happen because the internet was bad. It happened because the price was wrong. November gives Anthropic more time to explain why its price isn't.