Claude Code’s top 5% spend 20x a typical Enterprise user

Anthropic filed its S-1 with the SEC on October 26 to list on Nasdaq under the ticker ANT, and the prospectus makes Claude Code the center of its enterprise growth story. In the third quarter, the average Claude Enterprise user without Claude Code spent less than $75 a month, while the top 5% of Claude Code users on Enterprise spent about 20 times that. The multiple sets only the heaviest Claude Code users against an average seat without the tool, over a single quarter, so it shows the top end of spending rather than what a typical Claude Code seat costs.

The share count and price range haven’t been set, Anthropic’s announcement of the filing says. The growth figures come from the prospectus: as of September 30, 2026, Anthropic had more than 11,900 customers that paid it $100,000 or more over the trailing 12 months, up from more than 950 a year earlier.

Revenue next to the loss

Revenue reached $16.3 billion in the first half of 2026, up from $1.2 billion in the same stretch of 2025. Token usage accounted for 50% of it.

Readers will meet that revenue line next to a much larger number. The Wall Street Journal reported a net loss of more than $80 billion for the same first half of 2026. Both figures cover the same six months: one is what customers paid, the other is the bottom line after costs.

How Claude Code opens the door

The S-1 lays out a land-and-expand play, where a vendor wins a small foothold inside a company and then grows the account. An engineer picks up Claude Code, likes it and pushes it internally. By the time Anthropic’s sales team arrives, the filing says, “Claude is often already in use.” Anthropic calls Claude Code “the product we believe has most concretely defined what autonomous AI work looks like” and says non-technical teams are adopting it too, for a tool that started in the terminal and became generally available in May 2025.

“A model alone is not an agent,” the S-1 says. Anthropic sells the harness that runs Claude Code, the software around the model that drives its tools, to other developers. They can run it on their own servers through the Agent SDK or pay Anthropic to run it as Claude Managed Agents. That puts Anthropic closer to the companies building coding agents on top of its models. The S-1 names Google and OpenAI as its primary competitors in frontier models; Cursor, Replit and Cognition don’t appear in it, The New Stack notes.

Claude writing Anthropic’s code

Claude wrote more than 90% of the contributions to Anthropic’s codebase in September, the filing says, and the average engineer produced more than 45 times as much code per day in the third quarter as in 2024. Anthropic says its engineers increasingly direct and review Claude’s work. The New Stack points out that the company doesn’t define “contributions” or say how much human review that code gets.

That dependence also shows up in the risk factors. Anthropic lists Claude Code among the critical tools in its own development work, so problems with the product “could directly impair our ability to develop and deploy future models.” It also warns that as Claude takes on more engineering and research, the company “may lose visibility across certain processes in a way that may become difficult for any individual or group of humans to understand and control.”