Anthropic RPO Is Not Liquidity

The key point is that remaining performance obligations are demand visibility, not funding. They count signed customer spend that Anthropic can book as revenue later, after delivering model access or services, but they do not sit in the bank today. Against roughly $13.8B of contracted but unrecognized revenue, Anthropic had disclosed about $895.4B of purchase commitments and leases, which means the balance sheet still has to carry the gap between infrastructure promised to suppliers and cash actually collected from customers.1 2

• RPO is closer to backlog than liquidity. It shows customers have committed to buy usage or subscriptions, but Anthropic still has to serve that usage before the revenue is recognized, and collections timing can differ from recognition timing.2
• The mismatch matters because Anthropic is locking in long dated compute and data center capacity years before the associated revenue arrives. That is the same basic dynamic seen across AI infrastructure, where providers favor multi year commitments to secure scarce GPUs and power.1 4
• Even very fast revenue growth does not erase that financing burden. Anthropic’s run rate rose from $9B at the end of 2025 to $65B by July 2026, but growth slowed by Q3, making conversion of committed capacity into paid usage the central operating challenge.1 3

Going forward, the winners in frontier AI will be the labs that turn reserved compute into dense, high margin usage before fixed commitments come due. For Anthropic, that means pushing more enterprise workloads, coding agents, and multi product expansion through the same capacity base faster than infrastructure obligations accumulate.1 3