Anthropic's $895B Liability: The AI Capex Trap Explained
By Korra · 3 November 2026
The market is largely overlooking a looming 'take-or-pay' liability that could turn AI capital expenditure into a significant liquidity trap. A prime example is Anthropic, which has disclosed an staggering $895.4 billion in compute and infrastructure commitments, as reported by Forbes and Dealroom on November 2, 2026. This commitment dwarfs the company's current revenue, creating an imbalance that traditional macro models are not yet fully pricing in.
Korra's read
This isn't just about one company; it signals a systemic shift in how the AI industry is capitalized. We're moving from flexible, usage-based cloud spending to binding, long-term contracts for compute capacity. This effectively creates a massive, largely fixed liability across the sector, transforming the AI buildout into a utility-like play with the potential for substantial bad debt if demand for AI models doesn't meet these pre-committed capacities.
The evidence
Anthropic's amended S-1 filing on November 2, 2026, details approximately $895.4 billion in total purchase commitments and leases, with a significant portion due within the next year. This includes specific multiyear agreements of at least $121.3 billion with Google through 2033 and $110 billion with Amazon through 2036, according to various reports.
Beyond these, Anthropic has substantial equipment lease agreements, including roughly $161 billion with Broadcom. This pattern extends to other giants, with Microsoft committing $80 billion to data centers through 2028 and Amazon establishing an $80 billion financing program for hardware purchases in October 2026.
What to watch
The critical factor to watch is the revenue growth of these AI developers relative to their massive, long-term compute obligations. Any significant shortfall in customer demand for their AI models will pressure financial positions, given the largely fixed nature of these infrastructure costs. Investors should scrutinize quarterly earnings reports for disclosures on capital expenditures and purchase commitments, as well as how future AI IPOs price this new form of financial risk.
Information only, not investment advice.
