Microsoft and Meta have sharply cut their internal use of Claude, according to reporting by The Information. Microsoft is trimming more than a third off Claude spending projected above $1 billion a year, while Meta has halved the number of people using Claude Code.
Key Takeaways
- Microsoft’s cloud division reportedly drops its monthly per employee budget from $100,000 to around $10,000.
- Meta went from roughly 60,000 Claude Code users to 30,000, after spending more than $105 million on the tool in 28 days.
- Both companies are pushing in-house tools just as Anthropic turns into a direct rival for their workplace software.
Have an AI Sum Up This Article
ChatGPTWhat Microsoft and Meta cut back
The numbers come from The Information and were picked up by The Decoder in its analysis of Anthropic’s shift from partner to competitor, published October 5. Together they show Microsoft and Meta reducing their reliance on Claude at the same moment, each with its own replacement tools.
At Microsoft, executives Scott Guthrie and Jay Parikh told teams to favor internal tools. Claude spending, projected above $1 billion a year, is coming down by more than a third. In the cloud division, the monthly budget per employee reportedly falls from $100,000 to about $10,000.
Microsoft staff are being steered toward GitHub Copilot and OpenAI’s models. It builds on a first retreat this spring, when Microsoft dropped Claude Code after burning through its AI budget, back then inside a single division. That earlier decision ended Claude Code licenses in the Experiences & Devices group, home to Windows, Microsoft 365 and Teams, by June 30 and moved those teams to GitHub Copilot CLI, after per engineer costs had climbed to between $500 and $2,000 a month.
At Meta, the cut shows up in headcount. The number of Claude Code users reportedly fell from about 60,000 to 30,000, driven by layoffs and by the push for two internal tools, Muse Code and MetaCode.
The size of the bill explains the turn. Meta reportedly spent more than $105 million on Claude Code in a single 28 day period. It had already introduced cost controls in June, around the time Meta banned Claude Code and Codex for its engineers. That June policy blocked both tools from generating test tasks and from code analysis, out of fear that output from rival models would leak into Meta’s own training data.

Why Microsoft and Meta are easing off
Cost is the stated reason on both sides. Agentic coding tools burn through huge numbers of tokens, a worry IT leaders have voiced for months, as we noted when the cost of AI coding agents started to alarm CIOs.
Meta set the frame back in June with Meta AI Gateway and its token budget plans, a dashboard that tracks each team’s AI spending in real time, ahead of per team token budgets announced for 2027. Halving Claude Code users follows the same logic.
Competition weighs as much as the bill. According to The Information, Claude Cowork and ChatGPT Work are increasingly turning into alternatives to Office, which puts Anthropic on Microsoft’s home turf. Yesterday’s supplier now sells products aimed at its own customers’ customers.
Meta, for its part, is building competing products to sell to businesses. Seen from Microsoft and Meta, paying hundreds of millions of dollars for a rival’s tool amounts to funding that rival’s growth, and The Decoder describes Anthropic as a company growing extremely fast and becoming a headache for incumbents.
Anthropic is pressing hard on large accounts right now. On October 2 it announced a $100 million program to train 10,000 engineers to deploy Claude by the end of 2027, with first cohorts drawn from Accenture, Deloitte, McKinsey and Morgan Stanley. The pullback by Microsoft and Meta lands in the middle of that enterprise push.
More articles on Horizon
- Mistral Large 4 Takes On Security Work Rivals Refuse
- OpenAI Launches Dots, Always-On ChatGPT Agents
- Gemini 4 Argon: Google Ships Its Most Powerful Model
What Microsoft and Meta’s move means for Anthropic
Developers inside the two companies feel it first. Microsoft engineers move toward GitHub Copilot and OpenAI’s models, Meta engineers toward Muse Code and MetaCode, the tools Microsoft and Meta now put front and center.
Anthropic loses two heavyweight customers. Microsoft and Meta rank among the largest known users of Claude Code, and a cut of more than a third on spending projected above $1 billion shows up in revenue. Other big accounts are heading the other way: Anthropic announced on October 1 that Barclays is scaling Claude across its operations and client service.
Other large companies get a two sided signal. Two of the best equipped AI buyers now judge Claude Code too expensive at scale and prefer in-house tools or an allied vendor. Any IT department negotiating with Anthropic now has two public examples to lean on when it talks price.
On the competitive front, OpenAI comes out ahead at Microsoft, its longtime partner, and GitHub Copilot picks up users. The fight is moving toward office software, where Anthropic is advancing with Claude Cowork now merged with chat in a single app, up against Microsoft 365 and ChatGPT Work.
What comes next depends on how Anthropic answers on pricing and volume. A dedicated offer for large accounts, or revised limits for massive deployments, would be the logical response to the pullback by Microsoft and Meta, two very visible exits.
Follow the story on Horizon.


