Microsoft has slashed its expected internal spending on Anthropic technology by more than one-third, according to a report published by Channel Insider citing The Information. Company leaders had previously anticipated spending at least $1 billion each year on Claude and other Anthropic technology for employee use. Meanwhile, Meta has seen its internal Claude Code usage drop from roughly 60,000 employees earlier in 2026 to around 30,000 more recently, though the report notes that workforce reductions contributed to some of that decline.
The spending cuts at Microsoft extend beyond Anthropic. According to the report, individual employees' monthly AI spending caps in the Cloud and AI division fell from $100,000 to roughly $10,000 in most cases. Leadership has been encouraging developers to use GitHub Copilot and preferred models, including OpenAI's. At Meta, the company's own coding tools have been gaining ground: MetaCode now has more than 30,000 internal users, while Muse Code has more than 6,000. On Sept. 28, Meta announced Meta Enterprise Platform, an initiative to bring products including Muse Code, the Muse agent, and Muse API to businesses and developers. According to the report, Meta used Claude while developing its consumer Muse AI agent before switching to its own models for launch.
Neither company appears to be walking away from Claude entirely, the report finds. Microsoft continues to offer Anthropic models through Microsoft Foundry, allowing customers to use Claude while Microsoft provides the Foundry experience, Azure infrastructure, and billing. The change centers on employee use, where AI costs and competing in-house tools are pushing both companies to rethink how much they rely on third-party models. The report also notes that Microsoft restricted employee use of a newer Claude model while its legal teams assessed Anthropic's data-retention terms, though the company's external Anthropic relationship remains intact.
For Microsoft partners, Claude remains part of the customer-facing AI mix, the report explains. The internal spending controls instead show how even large technology companies are becoming more selective about which models employees use and how much they spend on them. According to the report, enterprises may want access to several models while setting tighter rules around token consumption, approved use cases, sensitive data, and which workloads justify premium models. Microsoft's approach offers an example of that split: it can continue giving customers model choice through Foundry while directing its own employees toward GitHub Copilot and other preferred tools. The report recommends that partners and MSPs evaluate the coding tool, underlying model, and purchasing route separately, confirming subscription eligibility, setting spending limits, reviewing data-retention terms, and measuring whether the selected model improves the workflow before expanding a customer's AI deployment. The rising discipline around AI spending suggests that enterprises will increasingly treat model access as a managed resource rather than an open-ended entitlement, forcing technology providers to justify every dollar with measurable workflow gains.