Meta and Anthropic Negotiate $10 Billion Computing Power Lease Deal
Via Thestar, Techinasia, Interestingengineering, New York Times and TechCrunch
- •Meta and Anthropic are in preliminary discussions for a computing power lease deal worth up to $10 billion over two years, according to Reuters.
- •The deal could create a new business line for Meta and reduce investor concerns about its data center spending.
- •Anthropic recently secured computing capacity at xAI's Colossus 1 data center, broadening its infrastructure partnerships.
- •A separate $400 million chip-backed loan reported by TechCrunch points to a growing market for AI infrastructure financing deals.
- •GPU-backed lending and compute leasing are becoming standard mechanisms for funding AI development at scale.
What Happens Next
+ Show− Hide
- →Meta establishes a compute-leasing revenue stream that partially offsets its $30B+ annual AI capital expenditure, giving Wall Street a clearer path to ROI on data center spending and lifting pressure on the stock's capex narrative.
- →Anthropic diversifies its infrastructure dependencies away from Amazon (its largest investor and cloud provider), weakening AWS's leverage in future commercial negotiations and signaling that leading AI labs treat compute supply as a multi-vendor strategic priority.
- →GPU-backed lending and compute leasing deals accelerate the financialization of AI infrastructure, drawing institutional investors and specialty lenders into a new asset class built around GPU collateral and contracted compute capacity.
Near-term: Within 1-3 months, competing hyperscalers — particularly Google Cloud and Microsoft Azure — face pressure to offer comparable compute-leasing terms to frontier AI labs, as Anthropic's multi-provider strategy becomes the default negotiating posture for well-funded labs. Long-term: Over 2-5 years, hyperscaler business models bifurcate into traditional cloud services and dedicated AI compute leasing divisions, with the latter commanding premium margins and reshaping data center build-out priorities toward large-scale, single-tenant AI training clusters.