Why is a rocket company partnering with an AI editor?

SpaceX acquired xAI on February 2, 2026. The partnership has been announced as Cursor and SpaceX, though this is likely the xAI division of SpaceX working alongside Cursor.

What the partnership unlocks

Colossus is xAI’s training cluster. They state it was built in 122 days and doubled to 200,000 GPUs in another 92, with 194 PB/s of total memory bandwidth, 3.6 Tbps per server, and over an exabyte of storage. They describe it as the most powerful AI training system yet.

Cursor released Composer 1, 1.5, and 2, iterating rapidly and achieving what they describe as “frontier-level performance at a fraction of the cost of other models”. They say their bottleneck was compute, and the partnership, announced by SpaceX, directly addresses that.

Where Cursor sits

Cursor’s tab-complete model and agent experience are stronger than VSCode’s. The downside is their closed-source approach and lack of access to Microsoft’s first-party extension marketplace. Both of these are small gripes if Cursor meets your daily needs.

It’s my preference over VSCode for AI-driven development. Most of my work happens in Claude Code and Neovim, but I still reach for a traditional editor when Neovim feels too restrictive.

Cursor started as a Visual Studio Code fork in 2023 and has diverged enough that the lineage is now mostly historical.

What I’m watching for

The news to watch for is Cursor’s next model release. Compute was reported as their bottleneck. The next release will show what that bottleneck was actually costing them.

What I wonder about

One thing that hasn’t been made clear by either party’s announcements is a tight coupling of Cursor with xAI’s existing flagship model, Grok. I’d be interested to see what a “Grok built for Cursor” would look like. xAI has announced their next generation of coding tools are coming soon, so the timing may not be a coincidence.

I’m interested to see what happens when a frontier model is trained for a specific editor. The editor becomes the model’s most-used tool. Its APIs become second-nature. Tooling mistakes get rarer and the surface area of what the model can do gets larger. Anything a human can do in the editor, the model could do too, given the right training and the right hooks exposed. That’s a model worth waiting for, and it would be exciting if this partnership gave us that.