Friday, Sep 4 | --:--
Back to home

Meta in Talks to Lease AI Compute to Anthropic in a Potential $10B Deal

On July 17, 2026, The New York Times reported Meta is in early talks to lease data-center compute to Anthropic in a deal that could be worth as much as $10 billion over two years—about a third the size of Anthropic’s May SpaceX compute pact—with either side able to opt out early; no deal is signed.

Tech Insights Reporter 4 min read San Francisco, CA
Cover illustration for Meta in Talks to Lease AI Compute to Anthropic in a Potential $10B Deal

TLDR

The New York Times on July 17, 2026 reported that Meta is in early-stage talks to lease AI data-center capacity to Anthropic in a structure that could be worth as much as ~$10 billion over two years, with monthly payments and early exit rights for either side. Anthropic proposed the deal in June; Meta is considering it. Both companies declined to comment. Talks may not result in a deal.

Deal shape (as reported)

From the NYT exclusive (three people familiar with the discussions):

Item Detail
Buyer Anthropic (compute lessee)
Seller Meta (capacity from its AI data centers)
Notional scale Up to ~$10B over two years
Payment Monthly increments
Flexibility Either party can opt out early
Status Early talks — not signed
Comparison Roughly one-third the SpaceX / Anthropic compute pact ($45B over three years, ~$1.25B/month, May 2026)

Context: Meta’s AI capex narrative (Zuckerberg-era spending on the order of tens to ~$145B annual AI-related investment figures in contemporaneous NYT framing) faces investor pressure; leasing spare or dedicated capacity to a rival frontier lab would be a new commercial line, not only internal Muse training burn.

Why this story matters

Compute is the real balance sheet of frontier AI. A Meta→Anthropic lease—if closed—would blur the line between hyperscaler landlord and model competitor, while proving Anthropic will stack multi-vendor capacity (SpaceX, cloud, now possibly Meta) rather than depend on one landlord. Readers following IPO paths and model races need the financing and infra thread: who owns GPUs, who rents them, and at what monthly burn.

Sources

Prior Coverage

Earlier Times of AI reporting on this thread.

Scroll to continue reading