Anthropic is in early talks to lease artificial-intelligence computing power from Meta in a deal that could be worth as much as $10 billion over two years, according to The New York Times, with CNBC separately confirming the discussions. The talks are preliminary, and the companies declined to comment.
The deal on the table
Under the arrangement being discussed, Anthropic would pay Meta in monthly increments across a two-year term. People familiar with the matter cautioned that the terms remain fluid and that either party could walk away early — the hallmarks of a negotiation still in its opening rounds rather than a signed contract.
Meta as landlord
The striking part is the counterparty. Meta is one of Anthropic's rivals in frontier AI, yet it is also sitting on one of the largest compute buildouts in the industry. Renting slack capacity to a competitor would turn Meta's enormous capital spending into a revenue stream and cast it, unexpectedly, in the role of cloud landlord — the news nudged Meta's shares off their lows.
Anthropic's compute hunger
The talks underline how compute has become Anthropic's binding constraint. Weeks earlier, the company agreed to tap Elon Musk's SpaceX and its Colossus 1 data center in a contract reported at roughly $45 billion over three years — about $1.25 billion a month. Stacking a Meta deal on top would spread Anthropic's capacity across multiple suppliers rather than a single hyperscaler. The SpaceX arrangement was struck to expand capacity for Anthropic's paying subscribers; adding Meta would give it a second heavyweight source and lower the risk that any one provider's outage or renegotiation stalls its roadmap.
The IPO backdrop
All of this is unfolding as Anthropic lines up investor meetings for a possible public listing later in 2026. Locking in multi-year compute — even at eye-watering cost — gives the company a more predictable capacity runway to show prospective shareholders, and reduces the risk that a chip crunch stalls its model roadmap mid-race.