On November 12, the most successful AI coding tool ever built loses access to the models that helped make it. Not for missing a payment. Not for breaching a policy. For changing owners.
OpenAI announced on August 28 that it is terminating its contract to supply models to Cursor, following SpaceX’s acquisition of Cursor’s parent company, Anysphere. The stated reason is ownership and nothing else. “We cannot be confident that SpaceX will use our technology within our terms of service,” OpenAI wrote, “based on our experience with Elon Musk’s companies violating contracts.” The company pointed to Twitter’s breach of OpenAI’s terms after Musk acquired it, and to Musk’s admission under oath that xAI had violated them as well.
The contract had a change-of-control clause. OpenAI pulled it. Cursor’s developers have about eleven weeks.
The numbers that make this worth your attention
This is not a small vendor losing a small supplier. Cursor reached roughly $3 billion in annualized revenue by late April. SpaceX acquired Anysphere in an all-stock deal valuing it at $60 billion — the largest startup acquisition on record — with a $10 billion breakup fee and a $4 billion penalty attached to antitrust failure. Weeks before that deal was announced in June, Anysphere had reportedly been raising at $50 billion on its own.
OpenAI says the two companies had worked together for nearly four years.
Four years of partnership, three billion in revenue, and the biggest exit in startup history bought Cursor exactly no durable right to the inputs its product runs on. That is the part worth sitting with, and it has nothing to do with which billionaire you find more sympathetic.
Anthropic went the other direction
In the same week, Anthropic confirmed it is keeping Claude in Cursor and signaled it would increase compute allocation to the product.
Both companies were making a commercial decision about a customer that competes with them. OpenAI ships Codex. Anthropic ships Claude Code. Cursor competes with both. OpenAI decided denial was worth more than the revenue. Anthropic decided distribution was worth more than the denial. Neither call was about principle, and pretending otherwise gets you the wrong lesson.
The useful signal for buyers is that “we will not cut you off” has just become a differentiator a model provider can sell — and one you should now be pricing when you sign.
What Cursor got right by accident
Cursor survives this because it was never single-sourced. Users can switch the underlying model. The product was built so that a supplier could disappear without the product disappearing with it, and that architectural decision — made years before anyone imagined this scenario — is the only reason this is a migration story rather than an obituary.
Most teams I talk to have not made that decision. They wired one API deep into their product because it was faster, the SDK was better, and multi-model abstraction felt like premature engineering. It is a defensible call right up until the week it is not.
What I think
I have been on the pleasant side of this exact dynamic, and it taught me more than the unpleasant version would have.
I was mid-workshop, in development on a helpdesk agent, when the model we were building on was repriced underneath us. The change landed in our favor — dramatically. Cost per ticket fell from roughly fifteen cents to about two. Overnight, a business case that had been an argument became obvious, and a system we were still justifying became one nobody needed to justify.
Here is what I actually take from that. I did not negotiate that outcome. I did not forecast it. I had no contractual claim to it, and if the number had moved seven times in the other direction instead, I would have had exactly the same amount of control: none. The pricing model my project depended on was a decision made by people who had never heard of my project.
We are all very good at banking the upside of that and calling it a win. We are much worse at noticing that the win and the catastrophe come from the same property of the arrangement, which is that we do not own the input. I got a favorable repricing. Cursor got seventy-six days.
So the practical version, for anyone building on somebody else’s models:
Read the change-of-control clause before you read the pricing. If your provider can walk on an ownership change, you have a hidden ceiling on your own exit options that nobody in your funding conversations has priced.
Treat multi-model support as insurance, not architecture polish. The premium is real — an abstraction layer, evaluation harnesses per model, prompt portability work. Cursor paid that premium and it just saved them.
Assume your provider becomes your competitor. Every major lab now ships an agentic coding product. The question was never whether the conflict arrives.
For three years the application layer has been priced as though model access were a commodity input: abundant, fungible, cheapening on a curve. This week someone demonstrated it is a relationship, and relationships have termination clauses. Cursor is large enough to absorb it. The next company this happens to will not be.
If you are working through an AI initiative and want a second technical opinion, or you want to talk through any of this, connect with me on LinkedIn or see the production side at coreagenticgroup.com.
2 comments