The Pulse
Meta and Microsoft Cut Internal Claude Use as Bills Rise
Meta and Microsoft are reducing employees’ internal use of Anthropic’s Claude while steering staff toward other AI tools. The companies continue to use Claude in customer-facing products, even as they seek to lower internal costs.
AI.info Team · Written with AI ·
“Internally, shifting more workloads to OpenAI models helps us get greater value from our token investment.”
Jay Parikh, executive vice president of Microsoft’s CoreAI engineering group, wrote in an August memo to staff.
Claude’s internal footprint shrinks
Meta and Microsoft are asking employees to rely less on Anthropic’s Claude, as both companies promote their own AI tools and seek to control the cost of staff usage. The moves put pressure on a pair of Anthropic’s largest corporate customers to balance internal dependence on Claude against efforts to build competing products. The Information reported the changes on October 5, citing people familiar with the companies’ spending and internal data.
At Meta, about 30,000 employees recently used Claude Code, Anthropic’s coding assistant, down from about 60,000 earlier this year. Spring layoffs, which cut roughly 10% of Meta’s then 78,000-person workforce, explain part of the drop. The Information reported that a larger factor was Meta’s push to develop and promote its own coding tools and models.
Meta builds replacements while setting limits
Meta’s internal data showed more than 6,000 employee users for Muse Code, a Claude Code competitor built on Meta models, and more than 30,000 for MetaCode, a similar tool intended for internal use. Meta began testing Muse Code with outside customers in August. The company also used Claude while developing its Muse AI consumer agent, then switched to its own models for the final product, launched in September.
In a recent 28-day period, Meta spent more than $105 million on Claude Code, according to internal data reviewed by The Information. The figure comes against a broader backdrop of heavy AI spending: a June memo cited in the report said Meta was on pace to spend billions of dollars on internal AI tools this year, without breaking out the amount going to each provider.
Meta’s Claude use had climbed this spring as employees competed on an internal leaderboard called “Claudeonomics,” which ranked heavy AI-tool users. In June, leaders pledged tighter controls, including usage budgets, limits on token consumption and a central dashboard to monitor external AI tools. That effort now accompanies a larger shift toward Meta’s own software.
Microsoft redirects spending to Copilot and OpenAI
Microsoft executives had projected that internal use of Anthropic’s technology would cost at least $1 billion annually. The company has since lowered that projected figure by more than a third after asking staff to use less Claude and spend more time with Microsoft’s own tools, The Information reported, citing two people familiar with the spending.
Microsoft has also cut back on Claude Code subscriptions and directed employees toward GitHub Copilot, where users can choose models. Copilot’s “auto” mode routes most requests to OpenAI models by default, which costs Microsoft less because of its commercial relationship with OpenAI. In an August memo, Parikh said each division would manage a token budget and urged employees to shift more work toward OpenAI models.
In Microsoft’s cloud and AI group, which employs more than 60,000 people, each employee had previously been allowed to spend as much as $100,000 per month on AI models. The division recently lowered that limit to around $10,000 in most cases, according to a person familiar with the change. Microsoft’s spokesperson confirmed that OpenAI models had become the default for internal AI coding, while saying engineers could still opt for Claude.
Customer demand keeps the partnership alive
The internal retrenchment does not mean Microsoft is abandoning Anthropic’s models in products sold to customers. Spending to power Claude features for Microsoft customers is separate from the internal-use figures, and The Information reported that this customer-facing spending has risen as Copilot subscribers grew. Microsoft’s total payments to Anthropic have remained roughly flat from their highs earlier this year, the report said, as increased customer usage offset lower internal costs.
That distinction captures the tension in the companies’ relationship: Microsoft and Meta compete with Anthropic while also buying its models. Meta is building coding software and consumer AI around its own models; Microsoft is steering employee workloads toward OpenAI while continuing to use Claude in customer-facing products. For Microsoft staff in its cloud division, the change is a lower monthly allowance: it had been as high as $100,000 per employee and is now around $10,000 in most cases.