From Partner to Rival: Meta and Microsoft Slash Internal Claude Use as Anthropic Becomes the Competition

For most of the past two years, the enterprise AI story has been written in superlatives: bigger models, bigger contracts, bigger spending. On October 5, 2026, The Information published a report that quietly flips that script. Meta Platforms and Microsoft — two of Anthropic’s largest corporate customers — are actively scaling back their employees’ use of Claude, steering internal workloads toward their own in-house AI tools instead.

The details are unusually concrete for a story about corporate procurement. According to the report by Aaron Holmes, Jyoti Mann, and Kevin McLaughlin, Microsoft slashed the monthly per-employee budget for Claude in its cloud division from $100,000 to $10,000 — a 90 percent cut. Microsoft had previously projected more than $1 billion in spending on internal Claude use, a figure that has now been cut by more than a third. Meta, for its part, halved the number of employees using Claude Code, Anthropic’s agentic coding tool, as it pushes engineers toward its own alternatives.

The numbers behind the retreat

The headline figures deserve context, because they describe a pullback of a magnitude that would have been unthinkable a year ago:

• Microsoft’s cloud division: monthly Claude budget per employee cut from $100,000 to $10,000
• Microsoft’s projected internal Claude spend: over $1 billion, now reduced by more than a third
• Meta: Claude Code user base halved as internal alternatives gain traction
• Both companies: continuing to offer Claude through their cloud platforms even as internal use shrinks

This is not a story about Claude failing in the enterprise. Anthropic’s annualized revenue run rate reached roughly $65 billion by the end of July 2026, up from $47 billion earlier in the year — a growth curve that very few software companies in history have matched. Claude Code alone was doing more than $2.5 billion in annualized run-rate revenue as of February 2026, with customers spending over $100,000 per year growing sevenfold. Enterprise customers drive the vast majority of that revenue.

The retreat is a story about identity. Meta and Microsoft are no longer just customers of Anthropic — they are competitors building the same class of products.

From partner to competitor

The strategic logic is straightforward. Meta has Llama-family models and its own agentic coding tools. Microsoft has GitHub Copilot, its Foundry platform, and the whole MAI model family under Mustafa Suleyman. Every dollar and every hour of developer attention that flows to Claude inside these companies strengthens a rival’s moat.

The shift from “partner” to “competitor” happened gradually, then suddenly. Microsoft gave its engineers Claude Code licenses in December 2025. Engineers loved it — so much that usage exploded and the token bills followed. By May 2026, reports emerged that Microsoft had canceled the vast majority of its internal Claude Code licenses after billing hit roughly $2,000 per engineer per month and burned through annual AI budgets. The June 30, 2026 cancellation date in the Experiences and Devices division marked what many observers called the clearest enterprise-scale AI spending pullback of the year.

But cost was only half the story then, and it is even less of the story now. Fortune’s reporting in May noted that canceling Claude Code licenses would not affect Microsoft’s Foundry deal with Anthropic, which includes investing up to $5 billion in the company. Microsoft remains an investor and distribution partner even as it pulls back internally — a textbook case of “coopetition” resolving toward competition.

Why this matters for the AI industry

Three implications stand out.

First, the enterprise AI market is entering its consolidation phase. The era in which every big tech company happily bought the best-in-class tool regardless of origin is ending. As Microsoft and Meta field credible internal alternatives, procurement decisions become strategic rather than purely technical. That is a headwind for independent labs like Anthropic and OpenAI, whose valuations assume continued hyper-growth in enterprise adoption.

Second, agentic coding tools are the beachhead — and the battleground. Claude Code’s success inside engineering organizations made Anthropic indispensable to companies that are themselves racing to build agentic products. The pullback shows that indispensability can be temporary. When your best product is also your competitor’s most-studied artifact, every internal deployment doubles as reconnaissance for the rival’s roadmap.

Third, token economics still bite at scale. The original Microsoft retreat was driven by bills, not strategy. Token-based billing for agent workloads — where a single task can spawn hundreds of model calls — turns a per-seat software cost model into something closer to a variable cloud-compute cost. Companies that mastered cloud cost optimization are now applying the same discipline to AI spend, and “best tool wins” loses to “good-enough tool at controllable cost” more often than vendors would like.

Anthropic’s position

None of this makes Anthropic a loser in the conventional sense. A company growing from $47 billion to $65 billion in annualized revenue inside a few months, with 500+ business customers spending over $1 million each annually, is not in trouble. The New York City Council hearing on October 5 — where Anthropic executives testified alongside OpenAI, Google, and Meta — underscored that the company is now firmly in the tier of firms that governments treat as infrastructure.

But the Meta-Microsoft pullback does foreshadow a structural challenge: the largest AI companies are increasingly becoming each other’s customers and competitors simultaneously, and the customer half of that relationship decays first. OpenAI faces the same dynamic with Microsoft, its largest backer and increasingly its most aggressive competitor in enterprise AI. Anthropic’s deep partnerships with Google Cloud and Amazon Web Services carry the same latent tension.

What to watch

The next data points that will confirm or refute this trend:

1. Anthropic’s next revenue disclosure — whether the growth rate holds as mega-customers trim
2. Microsoft’s Foundry-Anthropic relationship — whether the $5 billion investment partnership survives rising internal friction
3. Meta’s internal coding-agent metrics — whether in-house tools actually match Claude Code’s developer satisfaction, or whether engineers quietly route around mandates
4. Claude Code pricing changes — whether Anthropic responds with enterprise-friendly pricing to blunt cost-based defections

The broader lesson extends beyond AI. When a supplier’s product becomes strategic to your own roadmap, the relationship’s clock starts ticking. Meta and Microsoft are not saying Claude is bad — their engineers’ adoption proved the opposite. They are saying that in the super-intelligence era, feeding a future rival is a cost that no procurement budget can rationalize.

For Anthropic, the task is now to prove that its remaining thousands of enterprise customers — the ones without rival models to protect — provide a durable enough base to sustain the most aggressive growth story in software history.