Elon Musk’s AI company, xAI, is reportedly in talks with Cursor to provide the startup with access to large-scale computing power. This new deal comes as the coding startup works to strengthen its position against rivals such as Anthropic in the fast-evolving AI coding space.
Cursor is preparing to train its next AI coding model, Composer 2.5, on xAI infrastructure using tens of thousands of graphics processing units (GPUs), Business Insider reported, citing people familiar with the matter. The move highlights how access to computing resources is becoming a key factor in the AI race, especially for startups developing advanced models. It shows the Grok chatbot developer’s growing role beyond model development as it positions itself as an AI infrastructure provider to external companies.Under the arrangement, xAI will effectively act as a cloud provider, renting out a portion of its GPU capacity to Cursor, the report claims. This setup will allow xAI to generate revenue from its existing infrastructure while continuing to build its own models. It may also help offset the costs of expanding and maintaining large-scale data centres while strengthening ties with a startup that has access to developer-focused training data.
How this deal will help xAI to monetise infrastructure amid expansion push
The partnership comes as xAI continues to expand its data centre footprint through the Colossus project. Last year, the company said that it had around 200,000 Nvidia GPUs and plans to scale that number to 1 million. Elon Musk has previously stated that access to greater compute power would give xAI an advantage over competitors such as OpenAI and Anthropic.At the same time, the company is working to improve its hardware efficiency.
In a recent memo to staff, xAI president Michael Nicolls said the company’s model FLOPs Utilization (MFU) was “embarrassingly low” at about 11%, with a target of reaching 50% in the coming months. For comparison, industry estimates suggest that large-scale AI training systems typically operate at 35%-45% MFU, according to Lambda AI.The deal with Cursor could help improve utilisation rates by putting unused compute capacity to work.
It also aligns xAI more closely with established cloud providers such as Amazon, Microsoft, and Google, which rent computing infrastructure to developers and enterprises, as well as newer players like CoreWeave and Lambda that focus on supplying GPUs for AI workloads.For Cursor, the access to xAI’s infrastructure comes at a time of increasing competition in AI-powered coding tools. The startup is reportedly in talks for a valuation of around $50 billion, according to Bloomberg, but faces pressure from companies like Anthropic and OpenAI, which are expanding their own coding assistant offerings.Cursor recently released Composer 2, a model designed to generate and edit code across larger projects. The system was built on an open-source model from the Chinese startup Moonshot AI and further trained on Cursor’s own developer data. The upcoming Composer 2.5 is expected to build on these capabilities, with the added benefit of larger-scale training enabled by xAI’s compute resources.The relationship between the two companies also reflects some overlap in talent.
Earlier this year, xAI hired two former Cursor product engineering leads, Andrew Milich and Jason Ginsburg, who now oversee xAI’s product team and report directly to Musk and to president Michael Nicolls, according to a previous Business Insider report.In the meantime, xAI has seen some recent changes in its infrastructure leadership. The firm’s infrastructure lead, Heinrich Küttler, left last week. Jake Palmer has taken the lead on physical infrastructure, and SpaceX’s Daniel Dueri is now in charge of compute infrastructure.The partnership highlights the importance of access to computing power, talent, and data in the competition to develop AI. As companies build more capable models, partnerships like this could help determine how quickly startups can scale and compete with larger players.
