What it is: Meta is in early-stage talks to lease computing power from its AI data centers to rival AI company Anthropic in a potential deal worth up to $10 billion over two years . Anthropic would run its Claude AI models on Meta's data center infrastructure, including facilities in Georgia
. The discussions were first reported by the New York Times on July 17, 2026, citing three people with knowledge of the matter
.
Key terms as reported:
The Meta-Anthropic talks directly echo SpaceX's recent compute-lease agreements. Multiple sources describe Meta's move as following "SpaceX's playbook" .
SpaceX's deals with Anthropic and Google:
The parallel: Both Meta and SpaceX are massive infrastructure builders with excess compute capacity from their own AI buildouts. Rather than letting that capacity sit idle, they are becoming "landlords" to Anthropic and other AI companies — turning surplus GPU clusters into recurring revenue streams . The Meta-Anthropic deal would make Meta a landlord to Anthropic's Claude, the same role SpaceX played with Anthropic
.
Meta's discussions with Anthropic are part of a much larger strategic pivot first reported by Bloomberg on July 1, 2026 .
The Meta Compute initiative:
The infrastructure buildout driving this strategy:
Meta spent $72.2 billion on capex in 2025 alone and plans even more in 2026 . The Alberta data center alone is a 1-gigawatt hyperscale facility
. With this massive capacity buildout, Meta is seeking to monetize any compute it doesn't use internally — treating its data centers as both an engine for its own AI and a revenue-generating asset
.
As of July 17–18, 2026:
Bottom line: The deal is real in that active negotiations are confirmed by multiple credible sources, but it is early-stage, not finalized, and could still collapse.