How has Meta become both a public philosophical rival and a deeply dependent financial partner of Anthropic in 2026, including its projected spending of up to $10 billion annually on Anthropic’s AI technologies despite Mark Zuckerberg’s criticism of leading AI labs’ centralizatio

Meta’s 2026 relationship with Anthropic is a strategic paradox: it publicly champions a more open, decentralized AI future against the concentrated power of frontier labs, yet its engineers rely heavily on Anthropic’s Claude and Meta may become one of Anthropic’s largest customers. At the same time, Meta is positioning its vast AI infrastructure as a revenue source for Anthropic and, potentially, an eventual IPO beneficiary.

• Philosophical rivalry: Zuckerberg has argued that leading AI labs seek to consolidate power while presenting an unusually dire view of the future—a critique widely read as aimed at Anthropic and its safety-focused peers. Meta instead frames its strategy around broadly distributing AI capability, including through open-weight models. 1

• Operational dependence: Internally, Meta employees’ use of Claude—particularly the coding product Claude Code—became sufficiently extensive and expensive that the company projected annual Anthropic spending could reach $10 billion. Meta consequently imposed usage controls, in part to contain token costs and in part to prevent competitors’ model outputs from contaminating Meta’s own training data. 1

• The hedge—MetaCode: Meta is developing its own coding assistant, MetaCode, to shift routine work away from external vendors, cut bills, and reduce vendor lock-in. That is not evidence that Claude lacks value; rather, it reflects the risk of allowing a strategic competitor to become embedded in engineers’ daily workflow. 1

• Why Anthropic has leverage: Anthropic reportedly exceeded a $65 billion annualized revenue run rate by the end of July—an annualized extrapolation, not recognized full-year revenue—and was reportedly considering a possible public listing later in 2026. Its rapid commercial growth makes access to Claude strategically important and makes Meta’s internal reliance more costly. 2

• The relationship runs in both directions: Meta and Anthropic were in preliminary discussions for Anthropic to lease as much as $10 billion in Meta data-center computing capacity over two years, paid monthly. For Meta, this would turn infrastructure built primarily for its own AI ambitions into a cloud-like revenue stream and diversify it beyond advertising. It was a reported discussion, not a finalized deal. 3

• Why Meta can pursue this: Meta raised its 2026 capital-expenditure guidance to $130–$145 billion, mainly for AI infrastructure. Selling spare or dedicated capacity to an AI lab would help monetize that extraordinary build-out, even if the tenant is also a model supplier and ideological rival. 4

The contradiction is therefore commercial rather than accidental: Meta is paying Anthropic because Claude currently offers capabilities its workforce values; it is building MetaCode because that dependence is strategically uncomfortable; and it is courting Anthropic as an infrastructure customer because Anthropic’s growth—and a prospective IPO—could make it profitable for Meta to support the very firm it warns could centralize AI power. This is closer to “coopetition” than alliance: each company is simultaneously supplier, customer, competitor, and potential financial counterparty.