$META and $MSFT Curtail Internal Anthropic Spending to Transition Engineers Toward In-House AI Tools

Meta Platforms ($META) and Microsoft ($MSFT), two of Anthropic's largest enterprise customers, are sharply cutting back employee reliance on Claude as they prioritize proprietary developer tooling and rein in runaway AI token costs, according to a report from The Information. Microsoft, which was previously pacing toward at least $1 billion in annualized internal Anthropic expenditure earlier this year, has slashed that run-rate by more than one-third.
At Meta, active Claude Code developers have roughly halved from approximately 60,000 earlier this year to 30,000, even as the company incurred over $105 million in Claude Code expenses across a recent 28-day billing cycle. In parallel, Meta is redirecting internal workflows toward its own coding ecosystem, with internal assistant MetaCode exceeding 30,000 active users and its customer-facing model Muse Code scaling past 6,000 users.
The aggressive pullback from two anchor tech giants underscores escalating enterprise scrutiny over third-party API spend, highlighting how hyperscalers are leveraging in-house model architectures to capture developer mindshare while exerting direct margin pressure on frontier AI labs.