Anthropic reported third-quarter 2026 revenue of $16.8 billion, up 45% on the previous quarter. Then it pushed back its IPO from October to November. The bottom line: strong growth did not settle the question investors are asking about cost.
The Gartner gap
By the numbers, Gartner’s forecast is the sharpest. The firm sees at most $3.6 trillion in AI data center revenue through 2030. Covering planned investment, financing costs and a 12% return would take $8.5 trillion. Spending is running well ahead of money coming in.
What the prospectus shows
Anthropic’s leaked prospectus puts figures on that. Revenue for 2025 was $4.6 billion. Compute and infrastructure cost $7.33 billion that year, more than the company took in. The net loss was $42 billion. Investors will note that $34 billion of it was a noncash charge tied to financing instruments that could convert into shares.
The commitments ahead are larger. Anthropic expects to spend at least $518 billion over a decade on AI infrastructure with six partners, Reuters reported after reviewing the confidential prospectus. About 80% is noncancelable or payable whatever the usage. The company still targets a $2 trillion valuation in November, about 435 times its 2025 revenue. And a quarter of that 2025 revenue came from just two unnamed customers.
Warnings from OpenAI and New York
Anthropic is not alone in hesitating. OpenAI has said it will not list in 2026, and CEO Sam Altman called an IPO now “ill-advised” given AI safety concerns. Part of that is positioning. It is still notable when the head of the second-most valuable AI company chooses to wait.
New York City Comptroller Mark Levine went further and wrote to Anthropic CEO Dario Amodei, urging a pause. The city’s pension funds hold roughly $7 million in Anthropic exposure today, and could hold far more if the company joins the Russell 3000 after listing. Levine pointed to risks Anthropic itself flags: increasingly autonomous AI systems that could behave unexpectedly, and, in the prospectus’s own words, “existential risks to humanity.”
Open-weight pressure and debt
The figure that matters most for the valuation is the price gap. Enterprise adoption of open-weight models rose from 23% to 67% in 2026, with companies reporting savings of 70% to 90%. DeepSeek R1 offers 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 18 and 23 times higher. A $2 trillion valuation assumes closed models keep a durable premium, and that premium is being squeezed.
Debt is climbing too. Net debt plus future obligations for six major tech firms jumped from $576.66 billion to $3.61 trillion in two years. Only 13% of CIOs report significant value from their AI investments.
Sources
- The AI IPO Wave Is The Dot-Com Bubble's Evil Twin. Anthropic’s Delay Just Proved It.
- The AI Bubble Has A $3.6 Trillion Problem, And It's Getting Worse (NYSEARCA:SPY)
Photo: TechCrunch (CC BY 2.0)
