xAI Challenges CoreWeave in GPU Rentals

This makes xAI more than a model company, it makes xAI a merchant seller of frontier compute. The important shift is that Colossus is no longer just an internal training asset for Grok, it is a revenue generating GPU cluster sold to an outside lab that needs immediate capacity for Claude. That puts xAI and SpaceX into the same business CoreWeave built, leasing very large blocks of specialized AI infrastructure to customers that need thousands of accelerators fast.1 2 3 4

• Anthropic said on May 6, 2026 that it signed with SpaceX to use all of the compute capacity at Colossus 1, and xAI described Colossus 1 as a 220,000 plus GPU system for training, fine tuning, inference, and HPC. That is not overflow hosting, it is a hyperscale rental deal.1 2
• CoreWeave is the closest comp because its business is built around long term, large block GPU capacity deals. Its profile shows $5.1B estimated 2025 revenue and public market scale, while customer and partner documents describe bare metal GPU infrastructure and multiyear contracted capacity. xAI is entering that lane from a different starting point, an AI lab with captive demand first, external rentals second.3 5 6
• The field is segmenting by customer and product shape. Lambda sells cloud GPU access for training and fine tuning at scale, but with a more developer friendly reputation and smaller revenue base at $760M run rate in 2025. Together AI sits higher in the stack, wrapping leased compute in APIs and model hosting, and reached a $1B run rate by February 2026.4 7 8

The next step is a market where frontier labs, neoclouds, and infrastructure owners blur together. If xAI and SpaceX keep turning internal clusters into external capacity, the winners will be the groups that can secure power, chips, cooling, and financing fastest, then sell that capacity either raw like CoreWeave or bundled with models and software like Together AI.1 2 3 4 8