Meta and Microsoft sharply cut Claude usage as Anthropic becomes a rival

Microsoft cut the monthly per-employee budget for Claude in its cloud division from $100,000 to $10,000, and Meta halved its number of Claude Code users to 30,000, according to a report by The Decoder. The two companies, described as two of Anthropic's biggest enterprise customers, are pushing their own AI tools instead — and for Anthropic, that shift is converting a heavy reliance on a few major clients into a strategic risk.

The numbers are stark on their own. A 90 percent cut to Microsoft's Claude budget and a 50 percent reduction in Meta's Claude Code seats represent the kind of demand contraction that few AI vendors could absorb without consequence. When both cuts come from the same category of customer — hyperscale tech companies with their own competing AI products — the pattern points to something structural rather than cyclical.

What changed between Anthropic and its largest customers?

The Decoder frames the shift with a single phrase: Anthropic has transformed from partner into competitor. Microsoft sells AI through its cloud division and markets its own models and copilots. Meta builds and open-weights its own frontier models and deploys AI across its consumer products. Neither company needs an external model supplier indefinitely — and both now have internal alternatives they are actively promoting to their own employees and customers.

That dynamic reverses the conventional picture of the enterprise AI market, in which cloud giants act primarily as distribution channels for independent labs. Microsoft, in particular, has historically positioned itself as an aggregator: its cloud customers can access models from multiple providers alongside its own. Budget decisions like the one reported here — a tenfold reduction — signal that internal tools are winning share inside Microsoft itself, not just in its customer base.

Meta's move is narrower but equally telling. Claude Code is Anthropic's coding agent, aimed squarely at developer productivity — a category where Meta has invested heavily in its own tooling. Halving Claude Code seats to 30,000 users, as The Decoder reports, means Meta is substituting its own AI tools for a product it previously deployed at scale.

Why does customer concentration matter for Anthropic?

The core issue The Decoder identifies is concentration risk. When a large share of enterprise revenue flows from a handful of very large clients, the loss — or even partial retrenchment — of any one of them moves the needle materially. Losing budget from two simultaneously compounds the problem.

The strategic risk is sharpened by who those clients are. Anthropic's biggest users are not neutral buyers; they are AI builders with the engineering capacity and the commercial incentive to replace a third-party model with a first-party one. Every dollar of improvement in their own models is a dollar of reason to spend less with Anthropic. The Decoder's framing captures this: the more successful Anthropic becomes as a model provider, the more directly it competes with the companies that were once its best customers.

This is a familiar pattern in platform markets. A supplier sells components to large integrators; the integrators learn the technology; then they internalize it and cut purchases. For an AI lab, that dynamic arrives faster than in older software markets, because model capabilities are broadly visible, talent circulates, and switching costs for large engineering organizations are lower than for typical enterprise software.

The competitive pressure also runs through the partnership layer. Enterprise AI distribution increasingly flows through cloud marketplaces and bundled offerings controlled by the same companies retrenching from Claude. A vendor that competes with its distribution channel faces a structural conflict that budget lines like Microsoft's $10,000 cap make concrete.

What does this mean for the enterprise AI market?

For buyers and observers, the reported cuts are a data point in a broader question: can independent model labs sustain their position when their largest customers are also their most capable competitors? Anthropic is not alone in facing this. Every frontier lab that sells to big tech firms navigates the same tension between revenue today and substitution tomorrow.

The stakes are commercial rather than technical. Anthropic's models remain widely used, and Claude Code has been one of the most visible products in the AI coding category. But The Decoder's report shows how quickly enterprise demand can reallocate when a customer's incentives change — Microsoft's per-employee budget in its cloud division fell by an order of magnitude, not at the margin.

For Anthropic, the reported retrenchment by Meta and Microsoft puts a premium on diversifying its customer base toward organizations that do not build competing models, and on products whose value cannot be trivially replicated by a hyperscaler's internal tooling. The company's reliance on a few major clients, as The Decoder notes, is now a strategic risk rather than a sign of strength — and how it responds will say a great deal about the durability of independent AI labs' enterprise franchises.