Anthropic Posts $17 Billion Quarterly Revenue and Warns of Self-Improving AI^[1][9]

Anthropic generated $33.24 billion in revenue over the nine months ended Sept. 30, roughly a 1,200% year-over-year jump, while posting a $104 billion net loss driven largely by convertible notes and derivative liabilities.^[3] In the third quarter alone, revenue rose about 1,127% from a year earlier to roughly $17 billion alongside a $23.29 billion net loss.^[1][2]

Enterprise buyers are fueling that growth, with Anthropic reporting a 900 percent net revenue retention rate and roughly three-quarters of its customers in the first half of 2026 spending at least $100,000 annually.^[8] Still, quarter-over-quarter revenue growth cooled to about 47% in the third quarter, down from 144% in the second quarter and 132% in the first.^[2]

According to The Information, third-quarter operating costs hit $17.5 billion, erasing the prior quarter's profit.^[4] Anthropic blamed the surge mainly on $5 billion for short-term compute capacity and $2 billion in employee donation matching.^[5] Longer term, its amended S-1 discloses about $895.4 billion in purchase commitments and leases, nearly all for infrastructure and compute.^[6] According to the New York Times, the IPO could raise roughly $100 billion, and the company has struck dozens of compute deals with Amazon, Google, SpaceX, chipmakers, and former crypto miners.^[7]

Beyond the financials, Anthropic warned in the filing that internal estimates for upcoming models suggest it may soon trigger its Responsible Scaling Policy threshold for recursive self-improvement (RSI), defined as dramatically accelerated progress driven by AI.^[9] Anthropic said it does not expect to safely let Claude autonomously design and build future frontier models without human oversight, and is weighing policy changes to keep humans in control.^[10]