Anthropic is in preliminary discussions to lease $10 billion in computing resources from Meta.

      Anthropic is reportedly in initial discussions for a roughly $10 billion computing deal with Meta, according to the New York Times. This follows Meta's investment of $145 billion in AI infrastructure as it develops a cloud business to monetize these resources. CNBC has also confirmed these early talks. The discussions come shortly after Anthropic announced a $1.25 billion monthly agreement with SpaceX to utilize the Nvidia GPUs from the Colossus 1 data center.

      This potential deal highlights that, despite approaching a valuation of $1 trillion, Anthropic struggles to scale its computing capacity quickly enough to meet the demand for its Claude model. Access to AI chips continues to be a limitation, leading the company to impose usage caps on its most advanced models. Additionally, Google is renting GPUs from SpaceX for $920 million each month and is limiting access to its Gemini AI model for Meta due to high demand.

      For Meta, partnering with Anthropic could mark the start of the cloud computing venture that CEO Mark Zuckerberg has hinted at for some time. In May, Zuckerberg mentioned Meta was exploring entry into the cloud market and, last October, revealed that companies frequently inquire about purchasing computing resources from them at a premium. Meta might allocate up to $145 billion for capital expenditures in 2026. Furthermore, Dave Brown, a former senior executive from AWS, is set to join Meta. Following a reduction of 8,000 jobs in May, Meta is redirecting billions towards AI infrastructure, and selling excess computing capacity to Anthropic could demonstrate the value of that investment to stakeholders.

      This arrangement would create a unique scenario: while Meta develops its own Llama models and competes with Claude, it would also serve as Anthropic’s infrastructure provider. However, the current computing shortage has led to such collaborations becoming more common. SpaceX supplies GPUs to both Anthropic and Google, while Google provides Gemini to Meta, despite rationing access. The competitive boundaries in the AI industry are becoming less defined as all parties require more computing resources than they can independently provide.