Meta and Microsoft are pulling back on Claude and nudging employees toward their own in-house AI tools instead.
Microsoft has asked workers to lean on Copilot rather than Claude, after internal Claude spending was reportedly on pace to top $1 billion this year. Meta has cut its Claude Code user base roughly in half, from about 60,000 earlier this year to 30,000, while its homegrown MetaCode tool has already passed 30,000 users of its own. Meta also added token limits and spending controls after its internal AI costs were set to run into the billions. Neither company has explained its reasoning publicly, but the pattern points to cost-cutting and a preference for owning the tools their engineers depend on.
This matters because it mirrors a trend already playing out with smaller AI buyers, who have been swapping pricier US models for cheaper Chinese alternatives when the quality gap is small enough to live with. For Anthropic, losing scale with two of tech's biggest employers is a real revenue problem, not just a branding one, especially layered on top of the US Department of Defense recently flagging the company as a supply-chain risk. Anthropic itself has acknowledged in IPO filings that incidents like this could cause real revenue losses or business disruption.
Building your own AI stack to dodge a vendor's bill is a very Big Tech move. The irony is that the vendor they are now stepping away from is the one whose name briefly became shorthand for the category.