The most popular AI coding tool in the world is about to have a new owner, and it is not a software company. On June 16, 2026, SpaceX exercised an option to acquire Anysphere — the maker of Cursor — in a $60 billion all-stock deal, the largest acquisition of a venture-backed startup ever recorded. The deal is expected to close in Q3 2026, pending regulatory approval, after which Cursor becomes a wholly-owned SpaceX subsidiary.

If your team writes code in Cursor, that sentence has a practical edge: the model reading your codebase is about to answer to a company that also builds its own frontier model. Here is the founder's read — what's confirmed, what's actually at stake, and the short audit to run before the deal closes.

The 30-second version#

This was never really about the editor#

Read the price tag as a compute-and-data deal and it snaps into focus. SpaceX, through xAI, is buying three things it can't easily make:

  1. A data flywheel. Millions of developers using Cursor produce a continuous stream of real, in-context coding sessions — arguably the single best training signal for a coding model. Reporting indicates Grok 4.5 was built on an xAI model fine-tuned on Cursor coding data. Owning Cursor turns a licensing question into an internal one.
  2. Compute leverage. Cursor gains access to xAI's Colossus supercluster in Memphis — the kind of capacity constraint that had capped its growth.
  3. Talent. With xAI's founding engineers largely gone by the end of Q1 2026, a 60-billion-dollar acqui-hire of a top AI-tooling team is also a rebuild.

None of that is sinister on its face. But every one of those motives points the same direction: your code is the asset.

Cursor's zero-data-retention promise wasn't a feature to enterprise buyers — it was the reason legal, security, and compliance signed off at all. A change of control puts that promise in a new set of hands.

What actually changes for you#

Two things matter, and both are about control rather than the editor's UX.

1. Who controls your code. Cursor's Privacy Mode operates under zero-data-retention (ZDR) agreements with model providers — your code is not supposed to be stored or trained on. That is true today. Two caveats make it a live issue: Privacy Mode is a setting, not guaranteed to be the default on every seat, and after close SpaceX becomes the data controller for all code and prompts flowing through the tool. The ZDR terms you rely on were signed with Anysphere; whether they survive a change of control — and whether future changes need your consent or arrive as a quiet policy update — is a question you should be asking in writing, before Q3.

2. Whether the model stays your choice. Cursor today routes to Claude, GPT, Gemini, and Grok. Under an owner that ships its own frontier model, the realistic risk isn't a blunt "Grok only" mandate — it's drift: Grok as the default, the best pricing, the first feature. If the model that reads your code is increasingly chosen for you, you're using a different product than the one you adopted. This is the same lock-in axis we've written about for coding environments generally: the workflow can be great and still quietly own you.

The audit to run this week#

You don't need to migrate. You need to convert an ambient risk into a set of answered questions. Ask Cursor — in writing — and file the answers:

Then do two cheap things internally: keep your workflow model-swappable (route through a gateway so the provider is a config value, not a rewrite — our buyer's guide to picking an LLM API without lock-in is the playbook), and pre-qualify one alternative so leaving is a planned decision, not a fire drill. The field that made Cursor look essential now protects you: GitHub Copilot (with open-weight models in its picker), Windsurf (now OpenAI-owned), Zed, AWS Kiro, or a plain editor on direct API access. And if your codebase genuinely cannot be allowed near anyone's training run, treat that as its own project — here's how to keep your source code out of model training.

The honest bottom line#

Cursor is still an excellent tool, and nothing about your contract changes until the deal closes. The failure mode here isn't using Cursor — it's standardizing so completely on a single vendor that its ownership becoming a rocket company is your problem to solve on their timeline. The deal is a reminder that a developer tool is a supply chain, and the thing flowing through it is your intellectual property. Treat model choice and data control as procurement terms, keep one foot free, and this becomes a decision you get to make calmly — which is the only kind worth making.