Anthropic’s IPO Filing Shows Claude Code Driving Enterprise AI Spending

Claude Code appears to be doing double duty for Anthropic: bringing developers into the company’s products and expanding enterprise spending once it reaches their organizations. In its Monday IPO filing, Anthropic says the top 5% of Claude Code users on Claude Enterprise spent about 20 times as much in the third quarter as the average Enterprise user who didn’t use Claude Code, whose monthly spend was below $75. That gap is striking, though it doesn’t show how much of the difference comes from Claude Code itself, how much comes from heavier use of Claude more broadly, or whether the highest spenders represent typical customers.
For Anthropic, the figure makes Claude Code more than a developer feature: it’s a possible route into larger company accounts. The filing offers a snapshot of that commercial argument, while leaving questions about how durable the spending pattern will be as customers expand their use.
A developer tool becomes an enterprise entry point
The filing describes a sales path that often starts without a sales team. An engineer adopts Claude Code, then introduces it inside the company; by the time Anthropic’s sales team arrives, “Claude is often already in use.” That sequence matters because it gives the product a chance to establish a foothold through day-to-day work before a company makes a broader purchasing decision. The company calls Claude Code “the product we believe has most concretely defined what autonomous AI work looks like.” It says use is spreading to nontechnical teams as well.
That adoption coincides with a sharp increase in large customer accounts. Anthropic reported more than 11,900 customers with at least $100,000 in trailing 12-month payments as of September 30, 2026, compared with more than 950 a year earlier. Revenue reached $16.3 billion in the first half of 2026, up from $1.2 billion in the first half of 2025. Token usage generated half of the first-half 2026 revenue. Those figures show both rapid growth and the importance of usage-based demand, though the filing doesn’t establish how much of the increase came specifically from Claude Code.
Here’s the thing: a developer tool can open an enterprise door, but wider adoption still has to justify its cost. If you’re working in this space, the more useful signal may be whether an initial team’s use spreads across functions and turns into sustained company spending. Anthropic says that’s happening; the filing’s account figures provide scale, but don’t break out how many customers followed that path.
Anthropic uses Claude inside its own engineering organization
The filing says Claude produced more than 90% of contributions to Anthropic’s codebase in September. The company also reports that its average engineer produced more than 45 times as much code per day in the third quarter as in 2024, as engineers increasingly direct and review Claude’s output. Taken at face value, those are substantial changes in how the company says its engineering work gets done.
But output is an incomplete measure of productivity. The filing doesn’t define what counts as a codebase “contribution” or explain the extent of human review. More generated code could mean more work completed, but the practical question is how much of it becomes maintained, production-ready software—and how much effort is needed to test, revise, and support it. The numbers here are underwhelming as evidence of quality because they don’t answer that question.
That distinction matters beyond Anthropic. Code generation can shift engineers’ work toward directing and checking systems, as the company describes, without making the need for judgment disappear. Faster production is useful only if review keeps pace and the resulting software remains understandable to the people responsible for it.
Beyond the model: selling the agent runtime
Anthropic’s filing presents the software that operates Claude Code as another product opportunity. “A model alone is not an agent,” it says. Developers can run Anthropic’s harness on their own servers using the Agent SDK, or pay for Claude Managed Agents, where Anthropic operates it. The distinction gives customers a choice about who runs the system, while creating a commercial role for the software and services around the model itself.
That puts the company in the market for tools built around coding agents, as well as frontier models. The New Stack has covered GitHub and Anthropic using their own agents for large Rust rewrites, and AWS’s open-source agent, which the company says costs 45% less than Claude Code and Codex. Anthropic names Google and OpenAI as its primary frontier-model competitors in the filing; Cursor, Replit, and Cognition aren’t listed. That omission doesn’t establish how Anthropic views those companies, but it does show which rivals the filing identifies in this category.
The runtime question is practical, too: customers need to decide how much control they want over the infrastructure that coordinates an agent’s work. Running a harness on their own servers and paying Anthropic to manage it are different operating choices. The filing outlines both paths without describing how customers divide between them.
Dependence brings its own operational questions
Anthropic also relies on Claude Code in its own development work. Its risk disclosures say problems with the tool could directly impair the company’s ability to develop and deploy future models. As Claude takes on more engineering and research, Anthropic warns it may lose visibility into some processes in ways that become difficult for people to understand and control. That’s a direct acknowledgment that growing capability can bring new dependencies inside the company building the systems.
That tension sits alongside the company’s public call to pace frontier AI development, a position Dario Amodei has said may have cost Anthropic an AI lead. The filing gives investors a concrete measure of Claude Code’s commercial reach, while leaving open how Anthropic will manage its growing dependence on the tool. (And this is the part most people overlook.) The company’s own productivity claims and its warnings about reduced visibility describe two sides of the same adoption story.
What the filing leaves open
Anthropic’s account of Claude Code makes a persuasive case that a product used by engineers can help introduce the company to enterprise buyers and support higher spending. The reported figures are notable, but they don’t show whether the top spenders are representative, how much of their bill is attributable to Claude Code, or what returns those customers see. Those are meaningful gaps in a filing that puts the tool near the center of Anthropic’s growth story.
And yet. The same product appears to be changing how Anthropic works internally, while creating a dependency the company says could affect future development. For customers, the question is whether the tool’s value extends beyond faster code production; for Anthropic, it’s whether growth and reliance can be managed together. A larger footprint brings opportunity, and more operational weight.
Anthropic plans to list on Nasdaq under the ticker ANT. The filing doesn’t provide a price range.
The post Claude Code’s top enterprise users outspend other Claude users 20 to 1, Anthropic’s IPO filing says appeared first on The New Stack.