Anthropic's S-1 is the first audited look at what agents do to a lab's business
Anthropic filed its S-1 on Monday for a Nasdaq listing under the ticker ANT. The headline numbers: $16.3 billion of revenue in the first half of 2026, up from $1.2 billion, an operating loss of $693 million, and net revenue retention of about 900% as of September 30. More than 11,900 customers spend over $100,000 a year, 1,900 spend over $1 million, and about 250 spend over $10 million. The net loss was $80.7 billion, and the FT says more than $72 billion of that comes from revaluing convertible notes, not from running the business.
The part that matters for this site is how much of that growth is agents. TechCrunch's read of the filing: more than half of $100K+ customers now use both the Claude platform and a first-party app like Claude Code or Cowork, up from about 30%. The top 5% of Claude Enterprise customers using Claude Code spend about 20 times as much per month as the average enterprise user who doesn't. Inside the company, "over 90% of the contributions to our codebase in September 2026 were written by Claude", and daily code volume per engineer is over 45 times the 2024 average. The glossary even defines "the agent loop".
The risk factors are the honest half. Anthropic warns that agentic offerings "consume significantly more compute per user interaction than single-turn API calls", so compute needs "could increase disproportionately relative to the associated revenue." And it says products like Claude Code and Cowork "require broad permissions and deep integrations", which raises the stakes of any security incident or unauthorized action. That is the same loop seen from two sides: agents are why retention is 900%, and agents are why margins and liability are hard to forecast.
Two smaller details. Compute commitments with SpaceXAI run up to $120.5 billion into 2029, about $3.1 billion a month at full delivery. And the run-rate revenue line in the filing is literally blank ("was $ as of October 31"), so ignore any run-rate figure you see quoted. WSJ reports a target of the week of November 16, raising around $100 billion at a $1.5 to $2 trillion valuation. On the same day OpenAI launched a model sold on cost per task, the S-1 shows why that number is now the whole fight.
Anthropic announcement: https://www.anthropic.com/news/s1-sec S-1 on EDGAR: https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm
← Back to all articles
The part that matters for this site is how much of that growth is agents. TechCrunch's read of the filing: more than half of $100K+ customers now use both the Claude platform and a first-party app like Claude Code or Cowork, up from about 30%. The top 5% of Claude Enterprise customers using Claude Code spend about 20 times as much per month as the average enterprise user who doesn't. Inside the company, "over 90% of the contributions to our codebase in September 2026 were written by Claude", and daily code volume per engineer is over 45 times the 2024 average. The glossary even defines "the agent loop".
The risk factors are the honest half. Anthropic warns that agentic offerings "consume significantly more compute per user interaction than single-turn API calls", so compute needs "could increase disproportionately relative to the associated revenue." And it says products like Claude Code and Cowork "require broad permissions and deep integrations", which raises the stakes of any security incident or unauthorized action. That is the same loop seen from two sides: agents are why retention is 900%, and agents are why margins and liability are hard to forecast.
Two smaller details. Compute commitments with SpaceXAI run up to $120.5 billion into 2029, about $3.1 billion a month at full delivery. And the run-rate revenue line in the filing is literally blank ("was $ as of October 31"), so ignore any run-rate figure you see quoted. WSJ reports a target of the week of November 16, raising around $100 billion at a $1.5 to $2 trillion valuation. On the same day OpenAI launched a model sold on cost per task, the S-1 shows why that number is now the whole fight.
Anthropic announcement: https://www.anthropic.com/news/s1-sec S-1 on EDGAR: https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm
Comments