SpaceX Bought Cursor. The Code Still Needs an Independent Reviewer.

SpaceX is buying Cursor. Anysphere, the company behind it, sold for $60 billion in stock on June 16, and the tool that writes a big share of the world's enterprise code now lives inside a company that also owns a frontier model and the machines to train it. That's the part worth sitting with. Whatever writes the code, something separate still has to check it before it merges.

The terms arrived four days after SpaceX's Nasdaq debut. A SpaceX shell, X67, merges into Anysphere, and Cursor becomes a wholly owned subsidiary. It's expected to close in Q3 2026, pending regulators. There's a $10 billion breakup fee, and another $4 billion on top if the deal gets blocked on antitrust grounds. SpaceX swallowed xAI back in February, and it plans to run Cursor on Colossus, its supercomputer in Memphis.

What Cursor actually sold was neutrality#

Cursor's pitch was that it sat above the models. It ran Claude, GPT, and its own Composer models side by side, with zero-data-retention deals covering customer code. That's the part now in question. IDC's Deepika Giri doesn't expect Cursor to stay model-neutral under the new owner, and says those zero-data-retention agreements with the model providers could get renegotiated. Gartner's Nitish Tyagi put it plainly: model access restrictions move faster than the products that depend on them. Nothing's changed yet. No new pricing, no new data terms, and Cursor keeps running on its own until the deal closes. But the incentive just flipped.

Forget the neutrality fight for a second#

The bigger shift is dumber than that. The generator, the model, and the compute all report to one owner now. When the same company writes your code, runs the model under it, and owns the machines it trains on, the one job that can't also sit with that company is deciding whether the code is correct, safe, and yours to ship. A reviewer who answers to the author isn't a reviewer.

And this isn't really about Cursor. It's what consolidation does to a stack. The logos change, the buyers change, and the team hitting merge is left with the same question every time: who's vouching for this code, and what do they get out of saying yes?

The review pile doesn't shrink#

None of this changes the Tuesday-morning reality. Tools write more code, faster, from fewer vendors. More of it hits review, and it falls to whoever merges. A quicker generator with a bigger supercomputer behind it just means more to check, not less.

Hyrax reads every pull request, whatever wrote it, and fixes what's broken in the repo's own conventions, gated on the tests. It catches issues before a pull request even exists. Developers stay the ones who review and merge. That's the whole point of a separate review layer: it doesn't work for the thing that wrote the code.

Codegen will keep consolidating. Review shouldn't.

Ship clean code.