Story
Nov 3, 2026
Anthropic’s IPO Dream Runs Into an $895 Billion Compute Tab
Anthropic’s new figures give optimists evidence of a rapidly scaling AI business, while sceptics see a company locking itself into infrastructure costs that even blockbuster revenue may struggle to absorb. The IPO question is no longer whether demand exists, but what that demand will cost to serve.
Anthropic’s first IPO prospectus, released last week, left investors looking at a fast-moving AI company through an outdated lens: it contained financials only through the first half of the year. Monday’s amended S-1 filled in the third quarter — and sharpened the central tension around the planned listing.1
The new numbers show a business expanding at startling speed. Third-quarter revenue reached roughly $17 billion, up from $11.6 billion in the prior quarter and $1.4 billion a year earlier; revenue for the first nine months hit $33.24 billion, about 1,200% above the comparable period. Yet the quarterly net loss was $23.29 billion, while the nine-month loss reached $104 billion, largely because of accounting charges tied to convertible notes and derivative liabilities.2
The more immediate operating concern is compute. Anthropic swung from a $974 million operating profit in the second quarter to a $567 million operating loss in the third, blaming in part $5 billion of short-term compute that “currently costs significantly more” than longer-term arrangements. Adjusted operating margin slipped to 15% from 19%, even as adjusted profitability remained positive.3
That helps explain the extraordinary hedge embedded in the filing: $895.4 billion of purchase commitments and leases, “substantially all” for compute and infrastructure. About $447.5 billion is expected by the end of 2030, including nearly $70 billion due in the 12 months after September 30, 2026.4
For supporters, the spending buys capacity for a product suite whose enterprise demand is still climbing. Anthropic ended September with $131.8 billion in cash and short-term investments, and one analyst argued that frontier models can command premium prices, making it rational to pay up to reach key capability thresholds first.5
For valuation sceptics, however, that is precisely the gamble. With a prospective $1.5 trillion to $2 trillion valuation, investors could pay $43 to $57 for every dollar of trailing sales. Jay Ritter of the University of Florida said a valuation above $1.4 trillion would require a credible route to at least $100 billion in annual after-tax profit — and warned that “a lot of things have to go right.”5