Meta and Microsoft are curtailing employee access to Anthropic's Claude AI coding tools, redirecting developers toward proprietary alternatives like GitHub Copilot, MetaCode, and Muse Code. The shift is about cost control and strategic positioning, not product dissatisfaction — both companies remain significant Anthropic customers on the commercial side. The numbers tell the story of a deliberate squeeze. Microsoft slashed per-employee monthly AI spending caps from $100,000 to roughly $10,000 across its cloud and AI division, collapsing what had been projected as a $1 billion-plus annual internal Anthropic spend by more than a third. These are ceilings, not actual expenditures, but the signal to engineers is unmistakable: use our tools, not theirs. At Meta, Claude Code's internal user base fell from approximately 60,000 to 30,000 — a 50% drop attributed primarily to the strategic pivot toward MetaCode (30,000+ users) and Muse Code (6,000+ users), not just to layoffs. Yet Meta simultaneously spent over $105 million on Claude Code in a single 28-day period, proving that reduced headcount doesn't mean reduced dependency. The relationship is being restructured, not severed. This is vertical integration playing out in real time. Both companies are simultaneously Anthropic's biggest customers and its competitors. They want to sell Claude to enterprise clients through their platforms while ensuring their own developer workforces are locked into proprietary toolchains. The customer-facing spigot stays open; the internal one narrows. The security dimension is non-trivial. Coding assistants — whether Claude Code, Copilot, or internal alternatives — all carry risks around automated access to sensitive files, commands, and credentials. Anthropic patched vulnerabilities allowing unauthorized execution and API key theft before public disclosure. Microsoft's own RoguePilot vulnerability demonstrated how malicious commands in a GitHub Issue could hijack repositories. Switching tools doesn't eliminate these risks; it merely relocates them. Anthropic's reported $65 billion annualized revenue run rate suggests the broader market isn't contracting. What's happening is more precise: the largest platform companies are internalizing the coding assistant layer the same way they internalized cloud infrastructure, search, and social graphs. Engineers lose optionality; Anthropic loses its most powerful internal advocates at two of the world's most important software companies. The pattern is familiar from every previous platform era. The moment a capability becomes essential infrastructure, the platform owner absorbs it. Independent tool providers get relegated to serving the customers the platform can't reach directly. Anthropic's commercial revenue may keep climbing, but its strategic influence inside Big Tech's engineering cultures is being deliberately curtailed.
