Claude is being rationed by two of its biggest enterprise customers. The Information reported on October 5, 2026, that Microsoft and Meta are both actively reducing how much their own employees use Anthropic's Claude. Microsoft had been on track to spend at least $1 billion this year on internal use of Anthropic's technology; that projection has since been cut by more than a third. At Meta, employees using Claude Code have fallen from about 60,000 earlier this year to roughly 30,000.
Microsoft: monthly caps cut to a tenth, staff steered to its own tools
Microsoft's move looks like internal rationing. Cloud and AI executives Scott Guthrie and Jay Parikh told staff to switch to GitHub Copilot and OpenAI's models, according to the report, and most employees in that division saw their monthly AI usage cap fall from $100,000 to about $10,000. A Microsoft spokesperson responded that the company has been steering staff toward its GitHub Copilot coding tool, while engineers can still choose other models.
Two numbers must not be merged: what was cut is Microsoft employees' own usage, not what customers buy. The same report notes that enterprise customers' spending on Anthropic models through Microsoft's platforms is still growing steadily. Reading this as Anthropic losing its market is the most common misreading of the story.
Meta: from 60,000 users to 30,000, with in-house tools taking over
Part of Meta's decline is natural, following layoffs of about 10% of its workforce this spring. The main driver, though, is the push toward its own tools: the internal coding tool MetaCode already has more than 30,000 users, and Muse Code, built on Meta's own models, has more than 6,000 internal users, with staff being guided to migrate. Another figure shows the cost pressure behind it: Meta spent more than $105 million on Claude Code alone in a single 28-day stretch, and the company had already begun tightening AI costs in June. Meta did not comment on the report.
The real story: the customers are also the competitors
What matters here is not Anthropic's short-term revenue but a change in how enterprise AI spending gets decided. When a company sells its own coding tools and models, buying a rival's internal tools at massive scale both costs money and builds the rival's habits and reputation inside its own engineering culture. Cost is the stated reason; competition is the half that is harder to say out loud, and it applies to Meta and Microsoft alike.
The wind is blowing beyond Anthropic. More large buyers are setting per-team token caps and reconciling bills monthly — the same direction as Cohere North 2's per-employee token limits, which we covered earlier. For ordinary teams, the practical lesson is concrete: work out which tasks genuinely need the most expensive model, give everything else a cheaper route, and do not let the bill arrive before the policy does.
