Elon Musk's xAI Powers Cursor with Massive Computing Power

The artificial intelligence company xAI, founded by Elon Musk, is set to provide part of its computing power to the coding startup Cursor. This collaboration will allow Cursor to utilize tens of thousands of xAI's GPUs to train its AI model, Composer 2.5.

This initiative marks a strategic turning point for xAI, which is looking to monetize its infrastructure by renting its resources to other companies. By doing so, xAI could not only generate additional revenue but also strengthen its relationships with innovative startups like Cursor, which possesses valuable coding data.

In the cloud computing sector, giants like Amazon, Microsoft, and Google already dominate the market by renting their computing power to numerous businesses. Emerging players such as CoreWeave and Lambda have also specialized in providing GPUs to AI model developers. Access to robust computing power has become a crucial issue in the competition for AI development.

It is worth noting that xAI and Cursor already have close ties. In March, Cursor hired Andrew Milich and Jason Ginsburg, two former product engineering leaders, who now oversee xAI's product team under the direct leadership of Musk and Michael Nicolls, president of xAI.

xAI, which is among the most ambitious companies in AI model development, boasts one of the largest networks of data centers. During a meeting last December, Musk claimed that xAI would surpass competitors like OpenAI and Anthropic due to its ability to train its models with superior computing power.

The rapid expansion of xAI over the past two years, under the Colossus project, has allowed the company to declare that it owns approximately 200,000 Nvidia GPUs last year, with the goal of reaching one million GPUs.

However, xAI's infrastructure team has recently undergone leadership changes. Heinrich Küttler, the former head of infrastructure, has left the company, and Jake Palmer has been promoted to a leadership position. Daniel Dueri from SpaceX has also joined the computing infrastructure team.

An internal memo from Nicolls revealed that xAI's FLOPs Utilization Model (MFUs), which measures GPU utilization efficiency, was at an "embarrassing" level of 11%. Nicolls has set a target of 50% for the coming months, while most large-scale AI training sits between 35% and 45% MFU according to Lambda AI.

Cursor, currently in discussions for a valuation of $50 billion, faces fierce competition from startups like Anthropic and OpenAI, which are heavily investing in the development of coding assistants. In March, Cursor launched Composer 2, a coding model based on an open-source model from Moonshot AI, which it refined with its own user data.