• Meta Platforms (META) has cut the number of employees authorized to use Anthropic’s Claude from around 60,000 to 30,000, according to The Information, in a move to rein in costs and limit competitive risks.
  • The reduction follows internal data showing massive token consumption—over 60 trillion in 30 days—and earlier restrictions on external coding tools within Meta’s Applied AI group.
  • The decision highlights the balancing act for big tech firms adopting third-party AI while developing in-house models, with implications for vendor relationships and enterprise AI governance.

Meta Tightens Grip on External AI Tools

Meta Platforms has sharply reduced the number of employees permitted to use Anthropic’s Claude, cutting authorized access from roughly 60,000 to 30,000, according to people familiar with the matter. The move, first reported by The Information, is part of an effort to control costs and mitigate competitive risks as the company pours billions into its own AI infrastructure.

Earlier this year, Meta actively encouraged employees to experiment with Claude and Claude Code, even hosting internal AI-training events. An employee-built dashboard reportedly logged more than 60 trillion tokens of usage in a single 30-day period across the workforce—a figure that underscored both the popularity of the tool and the potential expense. But by late June, Meta had imposed stricter limits on Claude Code and OpenAI Codex for engineers in its Applied AI organization, citing concerns over model distillation and intellectual-property leakage.

The latest cut suggests Meta is now applying those restrictions more broadly, prioritizing teams where external tools deliver clear value while pushing others toward internal alternatives. The company declined to comment on the specific numbers, but a spokesperson said Meta regularly reviews its AI tooling to ensure “security, efficiency, and strategic alignment.”

Financial Pressures Mount

The clampdown comes amid soaring AI-related capital expenditures. In Q2 2026, Meta’s revenue jumped 28% year over year to $60.8 billion, but operating income fell 8% to $18.8 billion and operating margin narrowed to 31%. The company recorded $2.4 billion in legal charges and $1.2 billion in severance tied to May layoffs that affected about 8,000 employees—roughly 10% of staff.

Meta has projected $130 billion to $145 billion in 2026 capital spending, mostly for AI data centers and proprietary models. Free cash flow fell sharply in the second quarter as capex surged, even as ad revenue remained robust. That squeeze gives management a strong incentive to scrutinize every line of the AI budget, including external subscriptions and token consumption, which can scale rapidly with employee usage.

“Token usage is not a perfect proxy for productivity, but it’s a real cost,” said an industry analyst who asked not to be named. “Meta is trying to find the sweet spot between innovation and discipline.”

Strategic Implications for Anthropic and Beyond

The reduction could mean fewer paid enterprise seats for Anthropic, though Meta’s prior heavy usage signals the commercial value of Claude in large organizations. For Meta, the move may help contain vendor lock-in and data-governance risks, especially as it develops competing models through its internal AI initiatives. Employees who relied on Claude for coding, research, and automation may face new approval gates and migration to less familiar in-house tools.

Investors are likely to view the decision through a dual lens: cost control is positive, but heavy reliance on a rival’s model also suggests Meta’s own tools are not yet fully meeting employee needs. The company’s push to own more of its AI stack—from data centers to coding agents—will be tested by whether it can match the quality and usability that made Claude so popular internally.

A Meta spokesperson said the company remains committed to providing employees with “the best tools for their work,” and that access decisions are made on a team-by-team basis. Anthropic did not respond to a request for comment.

Correction: An earlier version of this article misstated the number of employees affected by Meta’s May layoffs. It was approximately 8,000, not 10,000.