OpenAI has told SpaceX it will wind down the contract that supplies OpenAI models to the AI coding editor Cursor, with a proposed shutoff date of November 12, 2026. The trigger was SpaceX’s acquisition of Cursor maker Anysphere, a $60 billion all-stock deal completed on August 14 according to financial filings. OpenAI published the decision on its own newsroom on August 28, saying it could not be confident SpaceX would use its technology within its terms of service.
Cursor is one of the most widely used AI coding tools among independent developers, so the practical question for anyone who works in it is how much actually breaks. The answer, for now, appears to be less than the headline suggests.
OpenAI’s stated reasoning
The company was unusually direct about why. “We are making this choice because we cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk’s companies violating contracts,” OpenAI wrote in its announcement.
It cited two specific episodes: Twitter, now part of SpaceX, breaking the terms of an OpenAI contract after Musk’s acquisition, and Musk admitting under oath earlier this year that xAI, also now part of SpaceX, had violated OpenAI’s terms of service.
The mechanism is contractual rather than punitive in the loose sense. OpenAI’s custom agreement with Cursor contained a change-of-control clause giving it a limited window to cancel after new ownership takes over. SpaceX closed on August 14. OpenAI notified them roughly two weeks later, inside that window.
OpenAI did choose the longest runway the contract allowed rather than the shortest. “To maximize the time that developers can retain access to our models through Cursor, we are giving the maximum notice provided by our contract,” it said, calling the decision “incredibly tough” and noting it had worked with Cursor for nearly four years.
How much of Cursor this actually touches
Cursor chief executive Michael Truell responded on X, saying OpenAI models serve about 5% of Cursor user traffic and that the company is speaking with the OpenAI team to resolve it.
That figure is Cursor’s own and should be read as such. Most Cursor users default to Anthropic’s Claude models or Google’s, and both remain available. Anthropic signalled it is staying, with co-founder Tom Brown noting publicly that Cursor has been a trusted Anthropic partner since Sonnet 3.5.
So if you open Cursor tomorrow and you are on a Claude or Gemini model, nothing changes for you. If you specifically use GPT models inside Cursor, you have until November 12 as things stand, and you will need to pick a different model or a different editor after that.
What this says about building a workflow on somebody else’s plumbing
Nothing was wrong with Cursor’s product. Nothing was wrong with OpenAI’s models. No usage policy was broken by a single developer using either one. A capability that thousands of freelancers had folded into their daily workflow is being withdrawn because two large companies fell out over unrelated contract history and a corporate ownership change activated a clause nobody using the software had ever read.
That is the shape of platform risk in the AI tooling layer right now, and it is a fairly common shape. The tool you use is often a thin layer over models it does not own, licensed under agreements it does not fully control. When the licensing relationship changes, your workflow changes, and you get told after the fact.
For a developer in Pakistan or the Gulf billing international clients, that has an obvious cost. If you have marketed yourself around a specific stack, quoted delivery timelines assuming a specific tool, or trained a small team on one editor, a vendor dispute you had no part in becomes your scheduling problem and possibly your margin problem.
The practical response is not to avoid AI tooling, which would be silly. It is to stay portable. Keep your prompts, snippets and project rules in plain files rather than locked inside one editor’s proprietary settings. Learn the underlying workflow rather than one product’s buttons, so switching editors costs a weekend and not a quarter. Avoid quoting a client a price that only works if one specific model stays available at one specific rate. And when you list skills on a profile, describe the capability, not the brand, since the brand may not be the same one in eighteen months. Our breakdown of AI skills clients are actually paying for goes into which of those capabilities carry the most weight.
Where this goes next
Truell’s phrasing left the door open, and November 12 is a long way off in an industry that reprices itself monthly. A negotiated arrangement, a narrower licence, or an OpenAI reversal are all still on the table, and Cursor has an obvious interest in getting one.
There is also a wider pattern forming that is worth watching. Model providers are increasingly deciding who gets to build on top of them based on corporate alignment rather than technical fit, and the alliances are hardening: Anthropic and SpaceX now have a compute relationship, and OpenAI is walking away from a company SpaceX owns. If the model layer keeps sorting itself into camps, the tools sitting above it will get sorted too, and the developers using those tools will be sorted along with them without ever being asked.





