Meta and Microsoft scale back internal use of Anthropic’s Claude, report says

Microsoft reportedly cut its projected internal spend on Anthropic technology by more than one-third from an earlier estimate of at least $1 billion, while steering employees to its own AI tools; customer spending on Anthropic models through Microsoft platforms continues to grow. At Meta, Claude Code usage fell from about 60,000 employees to roughly 30,000, partly amid layoffs affecting about 10% of its 78,000-person workforce and a push toward in-house tools. Meta reports more than 6,000 internal users for Muse Code and over 30,000 for MetaCode.

Analysis

The signal is operational discipline, not evidence that model demand is weakening. For Microsoft, replacing employee use of an external model with first-party tools could modestly reduce costs, but the more important test is whether Azure and enterprise customers retain access to the models they want: model choice can support cloud-platform stickiness even when Microsoft standardizes its own workforce. For Meta, internal adoption is not yet proof of a commercial moat. Its tools could improve engineering throughput and reduce vendor dependence, but savings are real only if output and software quality hold up; layoffs also make user-count comparisons a weak productivity measure.

Second-order risk: both firms may be shifting costs rather than eliminating them. More internal use of proprietary models can raise inference, infrastructure, and model-development costs, while reducing external spend. That trade is attractive only if total cost per useful task falls. The report is not enough to infer a material earnings change, and the spending estimates and adoption figures should be verified independently.

Near term, limited standalone price signal. Over 1–3 months, watch earnings commentary on AI operating expense, developer productivity, Azure model consumption, and customer choice. Over 6–18 months, the differentiator is whether first-party tools deliver measurable productivity without weakening third-party model access or raising compute intensity. Thesis reverses if internal tool adoption grows but AI-related costs rise faster than productivity or cloud monetization.

Key Decisions for Investors

• No standalone trade on this report. Treat it as a prompt to monitor AI unit economics, not as evidence of a near-term earnings revision for META or MSFT.
• At the next MSFT results, check Azure growth and management commentary on third-party model consumption alongside AI infrastructure and operating costs. Persistent customer demand across models would support the platform thesis; weaker consumption or rising costs without monetization would undermine it.
• For META, monitor evidence of engineering output and AI-related expense rather than internal user counts alone. If productivity gains are substantiated while expense intensity improves, the in-house strategy becomes a positive operating-leverage catalyst; otherwise, adoption may simply reflect mandated substitution.
• Potential relative-value watch: consider long MSFT versus META only if subsequent disclosures show durable external AI-platform monetization at Microsoft and measurable cost or productivity gains at Meta are absent. Reassess if either company's earnings guidance or reported AI cost trends contradict that divergence.