Elon Musk's xAI plans to supply AI computing power to coding startup Cursor

Elon Musk's artificial intelligence venture, xAI, is set to leverage its extensive computing resources through a strategic partnership with the coding startup Cursor. This collaboration aims to enhance Cursor's AI coding model, Composer 2.5, by utilizing xAI's robust infrastructure, specifically tens of thousands of graphic processing units (GPUs). This new arrangement positions xAI as a cloud service provider, allowing it to monetize its vast GPU resources while continuing to develop its AI models. By renting out its computing power, xAI could offset the substantial costs associated with maintaining and expanding its data centers. The partnership with Cursor is particularly advantageous, as the startup possesses valuable coding data that could benefit both parties. In the competitive landscape of AI, access to computing power is becoming increasingly crucial. Major players like Amazon, Microsoft, and Google dominate the cloud market, renting out millions of GPUs to numerous companies. Newer entrants, such as CoreWeave and Lambda, are also carving out niches by supplying GPUs specifically for AI model development. This isn't the first collaboration between xAI and Cursor; earlier this year, Cursor brought on board two former product engineering leads, Andrew Milich and Jason Ginsburg, who now play key roles in overseeing xAI’s product team, reporting directly to Musk and xAI’s president, Michael Nicolls. As xAI races to establish itself as a leader in AI model development, it has been rapidly expanding its data center operations, a project codenamed Colossus. Musk previously claimed that xAI's increased access to computing power would give it an edge over competitors like OpenAI and Anthropic. Currently, the company boasts around 200,000 Nvidia GPUs, with plans to scale up to 1 million. However, xAI's infrastructure team has experienced some turmoil recently, with key leadership changes. Following the departure of infrastructure lead Heinrich Küttler, Jake Palmer has stepped into a leadership role, and Daniel Dueri from SpaceX has taken charge of the compute infrastructure team. In a recent memo, Nicolls noted that the efficiency of GPU usage during AI training at xAI was alarmingly low at approximately 11%. He set a goal for the team to increase this metric to 50% within the next few months, as most large-scale AI training typically operates between 35% to 45% efficiency, according to Lambda AI. Meanwhile, Cursor is reportedly negotiating a valuation of around $50 billion, while facing stiff competition from other AI startups, including Anthropic and OpenAI, which are actively developing coding assistants. In March, Cursor launched Composer 2, an innovative coding model designed to generate and edit code for extensive projects, built upon an open-source AI model from the Chinese startup Moonshot AI.