Meta and Microsoft Are Quietly Cutting Back on Anthropic’s Claude

October 5, 2026

Two of the biggest names in Anthropic’s orbit are pulling back. Microsoft has cut its projected internal spending on Anthropic’s Claude models by more than a third, and Meta’s internal Claude Code user base has roughly halved, according to a report by The Information published October 5. Both companies are steering employees toward their own AI tools instead — a notable headwind for Anthropic as it prepares for its anticipated IPO.

What happened

According to The Information’s reporting, Microsoft executives earlier this year anticipated spending at least $1 billion annually on internal staff usage of Anthropic models. That projected run rate has since been reduced by more than a third, with company leadership introducing stricter token budgets and steering developers toward Microsoft’s homegrown infrastructure and OpenAI models via GitHub Copilot. Within the Cloud and AI division, individual monthly AI usage caps were reportedly cut from $100,000 down to roughly $10,000 in most cases. Importantly, the pullback is internal: Microsoft continues to use Anthropic models to power customer-facing Copilot features.

At Meta, internal usage of Anthropic’s coding assistant Claude Code fell from roughly 60,000 users earlier this year to around 30,000 recently. Workforce reductions contributed to part of the decline, but the primary driver has been Meta’s active push to transition engineers to its own tools: MetaCode, which now has over 30,000 internal users, and Muse Code, a Claude Code alternative built on Meta’s proprietary models. Meta also reportedly used Claude during development of its consumer Muse AI agent, then switched to its own internal models for the official launch.

Why now

The stated reasons are cost and data privacy. Running frontier models for tens of thousands of employees is expensive, and a range of enterprise clients are re-evaluating heavy reliance on premium models — some moving to lower-cost alternatives, open-source models, or renegotiated terms. The timing matters: Anthropic is preparing for a public debut that could value it at $2 trillion, and investors will be watching whether its biggest enterprise customers keep paying top dollar. So far the signal is mixed — The Information notes Anthropic has recorded strong overall revenue growth, significantly expanding its base of enterprise customers spending $100,000 or more annually.

Why it matters

This is the hyperscaler-vendor paradox in plain view. Amazon and Google are major Anthropic investors, Microsoft sells Anthropic-powered features to its customers, and all of them are also Anthropic’s competitors for enterprise AI budgets. When your biggest customers, investors, and rivals are the same handful of companies, loyalty has a shelf life — especially when per-seat model costs add up at 30,000-user scale. For Anthropic, the read is twofold: its models are still good enough that customers pay a premium to serve them to end users, but internal-seat economics face real pressure from cost-conscious buyers with in-house alternatives. And for the rest of the industry, expect this pattern to spread: as coding agents get cheaper to run and smaller models get better, the “buy Claude seats for everyone” era is giving way to a build-it-or-rent-cheaper calculus.

FAQ

Is Microsoft dropping Claude entirely?

No. The cuts target internal employee usage; Microsoft continues to use Anthropic models in customer-facing Copilot features.

Why is Meta pulling back?

Cost control and data considerations, plus its own maturing alternatives — MetaCode (30,000+ internal users) and Muse Code, built on Meta’s proprietary models.

Does this hurt Anthropic’s IPO?

It’s a headwind worth watching, but Anthropic’s overall revenue growth and expansion among large enterprise customers remain strong so far.

Who reported this?

The Information, with the report summarized by Stocktwits on October 5, 2026.

Sources: Stocktwits (citing The Information), Meta, Microsoft