Product1 distinct publisher3 min readPublished
OpenAI says it cannot be confident SpaceX will honour its terms, so one of its largest customers loses access. For anyone reselling model capacity, that puts an ownership-change risk in the stack no clause covers.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Somewhere on Monday a developer opens the model picker in Cursor and finds the OpenAI entries on their way out, through no fault of their own and with nothing they can do about it.
The two sides are sizing the breakup in different units, which is why neither number refutes the other. Cursor's Michael Truell put OpenAI at about 5% of Cursor user traffic, and OpenAI's head of core products, Thibault Sottiaux, answered that token usage is "not a proxy for revenue nor value created" and asked him to share the math [8][9]. Sottiaux is right about the mechanics, whatever his motive for saying so. Traffic share counts requests. Revenue counts what somebody pays for the hard requests. Teams tend to assume usage spreads evenly across the options in the dropdown; in practice users leave the cheap default in place for autocomplete, then reach for the expensive model when the task is frightening. A small share of calls can carry a large share of spend, and Cursor's place in OpenAI's top five customers by revenue suggests it did [4].
The arithmetic on OpenAI's side is easy and worth doing out loud. More than $1 billion in projected annualized revenue against a reported run rate above $40 billion is roughly 2.5% of the top line [5][10][14]. Affordable, and also specific: OpenAI has now published a price for keeping Musk's companies away from its weights, and every other partner in its book can read it.
The trust rationale points somewhere particular. OpenAI's post noted that Musk appeared to say in a deposition, in a suit a federal jury dismissed earlier this year, that xAI, now part of SpaceX, used OpenAI's models to train its own [15]. Set beside the wind-down, the worry is distillation by a new parent rather than anything Cursor's users did. The two firms had been selling against each other for well over a year, with OpenAI building Codex into the same market [13], and OpenAI's own startup fund sat in Cursor's 2023 seed round [12]. Ownership changed, and the commercial logic that had governed the relationship no longer applied [2].
That is a forcing function for anyone whose product resells someone else's model. Two cells, filled honestly. First: what a swap costs, measured as engineering days plus the difference in retention and task completion you observed while actually running a week on the fallback, not the difference you assume. Second: which event outside your control ends the contract, with the acquirer or the litigant named. If the first cell reads "we have never run on the fallback" and the second cell has a name in it, that dependency is political. The master agreement will not cure it. OpenAI conceded the decision may cost it standing with developers [6]; the developers it costs are the ones who read multi-model support as insurance [11].
Ranked by verification strength, evidence, and original report placement.
In a late-night blog post last Friday, OpenAI said it would wind down its partnership with Cursor, the startup behind one of the most popular AI coding tools on the market.
OpenAI said in its blog post: "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."
At the start of 2026, Cursor was one of OpenAI's top five customers in terms of revenue, according to people familiar with the matter.
By the spring of this year, OpenAI estimated Cursor would bring in more than $1 billion in annualized revenue for OpenAI, based on the partnership's performance at that time.
SpaceX recently acquired Cursor in a $60 billion deal.
OpenAI acknowledged in its blog post that ending the partnership may damage its standing with developers by cutting off a major channel they use to access OpenAI's models.
Distinct publishers with included, body-backed reporting in this cluster.
2 articles · September 3, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
build
OpenAI's Cursor cutoff reclassifies model access as a supply-chain dependency8 distinct publishers
product
Termination notice targets Cursor's contract for OpenAI models1 distinct publisher
product
SpaceX paid $60B for Cursor to fix Grok, and Cursor's customers inherited the plan1 distinct publisher
product
Cursor's case for being bought is unit economics, and the post-close note promises nothing else4 distinct publishers
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One newsroom, one anonymous spine
The quotable layer is solid: OpenAI's blog language, Truell's post, Sottiaux's reply, the dismissed lawsuit. The numbers that make the story a story — top-five customer, $1B+ annualized — rest entirely on people familiar with the matter speaking to WIRED, which says outright they have not been reported before. OpenAI declined to comment; SpaceX and Cursor did not reply. The story appears twice in our coverage, but that is the same piece twice, not corroboration.
Cutoff announced, impact unmeasured
A real commercial action with a date attached — and almost nothing that measures its consequences. Truell's 5 percent of traffic and OpenAI's revenue framing are not the same metric, and neither side has published a number that would settle it. The only observable continuity is Anthropic keeping Claude inside Cursor, which WIRED reads as a favour owed to a landlord holding $45 billion of data center capacity rather than a verdict on the product.
A projection carrying a headline
"Billion-dollar customer" was OpenAI's own spring extrapolation from how the partnership was performing then, not revenue booked or lost, and against a reported $40B+ run rate it works out to roughly 2.5 percent. The trust and distillation story is the larger one; the money framing is stretched a little further than the arithmetic supports, and WIRED itself concedes the loss may not hurt as much as it once would have.
No neutral narrators left
Count the stakes. OpenAI is assembling a pre-IPO story about not depending on Musk's goodwill, and its startup fund seeded Cursor in 2023 while Codex now competes with it. Truell is minimising the loss from inside SpaceX. Anthropic keeps Claude in Cursor while owing SpaceX $45 billion of data center capacity, a year after cutting off Windsurf over a rumoured OpenAI acquisition. Even the distillation argument doubles as litigation positioning against a suit a jury already dismissed.
Trustworthy telling, unverified core
The public record part of this — who said what, when, and the dismissed lawsuit — can be checked and is unlikely to move. The financial spine cannot be checked from outside WIRED's sourcing, and the two sides' magnitude claims are measured in different units on purpose. Enough to act on the direction of travel; not enough to put a number on the damage.