Elon Musk confirms SpaceX will hand over access to the 220,000-GPU Colossus 1 supercomputer in Memphis, while xAI shifts its training workloads to the newer Colossus 2.
SpaceX and Anthropic have struck a deal that gives the Claude maker exclusive access to all compute capacity on the Colossus 1 AI supercomputer in Memphis, Tennessee. The lease runs for an initial six months with mutual 90-day cancellation rights, a structure Elon Musk says SpaceX itself requested.
The arrangement hands Anthropic more than 220,000 Nvidia GPUs, including H100, H200, and GB200 models, along with over 300 megawatts of power. That is a staggering amount of compute for inference workloads, the kind of processing that keeps chatbots responsive when millions of users hit them at the same time.
What Anthropic gets, and what it means for Claude #
Colossus 1’s firepower is aimed squarely at Anthropic’s inference stack, making Claude faster and more available rather than training the next frontier model from scratch.
The direct beneficiaries are Claude Pro and Claude Max subscribers, who can expect increased service limits and computational headroom.
Meanwhile, xAI has migrated its Grok model training to the newer Colossus 2 system. Anthropic gets immediate access to a massive GPU cluster it didn’t have to build, and xAI frees itself to focus training runs on next-generation hardware without competing for cycles on the same machines.
The business structure behind the handshake #
Musk confirmed on May 28, 2026, that the lease is structured as a 180-day agreement. Either party can walk away with 90 days’ notice, which effectively means there is no real lock-in beyond the first three months.
Early reports floated potential multi-year financial implications reaching up to $15 billion annually through 2029. Musk pushed back on those figures, emphasizing the short-term nature of the deal and the fact that SpaceX initiated the arrangement.
Reports indicate Anthropic is exploring orbital AI compute development in collaboration with SpaceX, establishing a working commercial relationship between the two companies.
Why this matters beyond AI #
Anthropic’s move to lease rather than build is instructive. Constructing a 220,000-GPU cluster from scratch would take years and billions of dollars. Leasing one that already exists and happens to be sitting idle because its owner moved to newer hardware is a shortcut that few companies are positioned to take.
The competitive implications for Anthropic’s rivals are real. Google, OpenAI, and Meta all operate their own massive compute clusters. Anthropic has historically relied on partnerships, most notably with Amazon Web Services, for its infrastructure needs. Adding Colossus 1 to its arsenal, even temporarily, gives it a meaningful boost in raw inference capacity.
The fact that no cryptocurrency or digital asset component is attached to this deal is notable. This is a straightforward compute infrastructure transaction with no tokens, no blockchain integration, and no speculative layer.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our