Elon Musk’s XAI brings compute to cursors for training AI models

AI For Business


Elon Musk’s AI company xAI plans to leverage its accumulated computing power in a new deal with coding startup Cursor, people familiar with the matter said.

Officials said Cursor plans to train its latest AI coding model, Composer 2.5, on the xAI infrastructure. Cursor plans to use tens of thousands of xAI’s graphics processing units (GPUs), the chips used to train AI models.

This setup effectively turns xAI into a cloud provider of sorts. By renting some of its GPUs to other companies, xAI can start monetizing its large infrastructure while developing its own AI models. The deal could help the company offset the costs of building and operating data centers, while also deepening its relationships with startups that have access to valuable coding data.

The biggest cloud providers, Amazon, Microsoft and Google, own millions of chips and rent out their computing power to thousands of companies and developers, generating huge profits. New players like CoreWeave and Lambda are building their businesses around supplying GPUs to AI model developers. Access to computing power is an increasingly important competitive aspect of the AI ​​arms race.

Representatives for xAI and Cursor did not respond to requests for comment.

This isn’t the first time Cursor and xAI have overlapped. The company hired two former Cursor product engineering leads, Andrew Milich and Jason Pinsburg, in March. Business Insider previously reported that Ginsburg and Milic oversee xAI’s product teams and report directly to Musk and xAI president Michael Nichols.

xAI is one of many companies competing to build the best AI models and has one of the largest data center footprints. Musk said at an all-hands meeting last December that xAI will beat competitors like OpenAI and Anthropic because it has more power available to train models.

Over the past two years, xAI has rapidly expanded its data center footprint, a project it calls Colossus. The company said last year that it had about 200,000 Nvidia GPUs, but Musk said it had plans to expand to 1 million GPUs.

xAI’s infrastructure team is experiencing a leadership shakeup. Last week, the company lost its head of infrastructure, Heinrich Kütler. Business Insider previously reported that the company moved Jake Palmer to a leadership role on its physical infrastructure team, and last week SpaceX’s Daniel Dueli took a leadership position on its compute infrastructure team.

xAI President Nichols said in a memo to employees last week that the company’s model FLOPs utilization (MFU), which measures how efficiently GPUs are used during AI training, is “embarrassingly low” at about 11%. Nichols said the team’s goal is to reach a 50% winning percentage within the next few months. For comparison, most large-scale AI training runs at 35% to 45% MFU, according to AI infrastructure company Lambda AI.

Cursor is in talks for a valuation of about $50 billion, Bloomberg reported last month. Meanwhile, major AI startups like Anthropic and OpenAI are actively building coding assistants and are facing pressure.

In March, Cursor released Composer 2, a coding model designed to generate and edit code across large projects. Cursor built its model on an open source AI model from Chinese startup Moonshot AI and fine-tuned it using proprietary data from its developer user base.

Do you work for xAI and do you have a tip? Email this reporter at: gkay@businessinsider.com Or text us at 248-894-6012. Use a personal email address, non-work device, and non-work Wi-Fi. Here’s a guide to sharing your information securely.