OpenAI cuts off Cursor's models after SpaceX takeover
OpenAI is winding down the model-supply contract that powers Cursor after SpaceX completed its $60 billion acquisition of the coding tool, citing distrust in SpaceX's compliance record. For African developers who have spent three years building workflows around Cursor, the dispute is a live lesson in a risk they rarely get to negotiate away: the tools they build on can be switched off by decisions made an ocean away, for reasons that have nothing to do with them.
Five percent of Cursor's traffic runs on OpenAI's models. That is the number Cursor's chief executive, Michael Truell, gave when OpenAI notified his company on August 28 that it intends to stop supplying those models entirely, with a proposed cutoff of November 12.
The timing is not coincidental. SpaceX closed a $60 billion all-stock acquisition of Cursor's parent company, Anysphere, on August 14. Two weeks later, OpenAI told SpaceX it could not be confident the rocket company would keep using its technology within the terms of service, and pointed to its own history with Elon Musk's companies as the reason: Twitter, OpenAI says, broke a data-sharing agreement after Musk bought it in 2022, and Musk has acknowledged under oath that xAI trained Grok on outputs from OpenAI's models in violation of its terms. OpenAI's own account of the decision, published August 28, frames it as protecting how its technology is used by a company it no longer trusts to police itself. Reuters and CNBC independently confirmed the November 12 date and the traffic figure, both citing OpenAI's statement and Truell's own public remarks.
OpenAI has also said it will withhold future models, including its next flagship release, Astra, from Cursor for the duration of the wind-down. Anthropic, whose Claude models make up a much larger share of what Cursor's users actually run, has said the opposite: it intends to increase the compute it dedicates to Cursor. Coding assistants have become one of the most contested layers of the AI stack precisely because they sit between a developer and every line of code that developer writes, and every company that supplies a model into that layer wants to be the one still standing when a rival gets cut off.
None of this has, so far, been the subject of a single African-focused headline. It should be. GitHub's own Octoverse 2025 report, published in October, found that Africa and the Middle East together added roughly 3.4 million new developers to the platform in the year to August 2025, a growth rate the report's authors linked directly to AI-assisted tooling lowering the barrier to a first commit. Nearly 80 percent of new GitHub users now try an AI coding assistant within their first week on the platform. In Lagos, Nairobi and Accra, that assistant is very often Cursor: cheap enough at $20 a month to expense personally, familiar enough to anyone coming from Visual Studio Code, and, until last week, uncomplicated by the geopolitics of its ownership.
That uncomplicated period is over, and the lesson it leaves behind has nothing to do with Cursor specifically. An African startup building on any proprietary AI coding tool is, by construction, one acquisition or one contract dispute away from losing a capability it did not build and cannot replace on its own timeline. Cursor's own developer base will likely absorb the November cutoff without much drama, because five percent of traffic is a small enough share that Truell has said talks are ongoing and a fallback to Claude or Gemini inside the same editor is a menu click away. But the fact that the fix is that easy for Cursor's users is itself the point: the leverage sits entirely with the platform, not the developer, and an African founder locked into a single vendor's routing decisions has no comparable menu to fall back on if the vendor decides the relationship no longer serves it.
There is a narrower way to read this than as a geopolitics story, and it is the more useful one for a developer deciding what to build on next. Our sister site Macrostack, which compares paid developer tools against open-source and self-hostable alternatives, rates Cursor's biggest structural weakness as architectural rather than commercial: even a Cursor user supplying their own API key still has every request routed through Cursor's own backend, a design the company's own documentation confirms. Open alternatives such as Cline or opencode remove that dependency entirely, running as extensions inside an existing editor while letting a developer point requests directly at whichever model provider they choose, with no third party positioned to cut the line. That is a meaningfully different risk profile from the one Cursor's own users are living through this week, and it costs nothing to adopt beyond the model bill a developer would be paying either way.
What comes next for Cursor is a negotiation whose outcome nobody can predict: Truell has said talks with OpenAI continue, and a shortened or reversed cutoff is possible. What comes next for the African developers who never had a seat at that negotiation is a smaller, sharper question worth sitting with regardless of how it resolves: when the tool a business is built on can be reshaped by a boardroom decision three time zones away, how much of that business's future is actually in its own hands?
