Cursor plans to train its Composer 2.5 model on xAI's GPU infrastructure, turning xAI into a cloud provider

Cursor to Train Composer 2.5 on xAI's GPUs, Positioning xAI as Cloud Provider

Cursor will use tens of thousands of xAI's GPUs to train its Composer 2.5 code model, according to people familiar with the matter. The arrangement effectively converts xAI's data center capacity into a cloud provider operation, allowing the company to generate revenue from its infrastructure while continuing to develop its own AI models.

The deal surfaces a concrete answer to a structural problem xAI has faced since mid-2025: Elon Musk's team built out compute capacity faster than demand materialized. xAI spun up massive GPU capacity ahead of most timelines, but customer adoption lagged. Cursor, which has built significant demand for its coding products, needed its own model training infrastructure. The pairing solves both sides—xAI gets a paying customer and revenue stream to offset data center costs; Cursor gets the compute it needs.

There may also be a trade component. Cursor has access to valuable coding data from its user base, which could deepen xAI's competitive position in code generation while also offsetting the cost of supplying GPUs.

The move raises a broader question about how companies marshaling massive compute capacity will monetize unused infrastructure. Musk is pursuing similar logic at scale with SpaceX data centers and his TerraFab chip manufacturing ambitions—build capacity, then find external customers to fill it and defray costs. Whether that demand actually materializes remains unclear, but Cursor represents at least one concrete anchor.