A price the seller could carry
In a late-night blog post last Friday, OpenAI announced it would wind down its partnership with Cursor, one of the widely used code editors through which developers reach frontier models. The ground it gave is trust: after SpaceX bought Cursor in a deal worth 60 billion dollars, OpenAI said it cannot be confident that SpaceX will use its technology within its terms of service, citing its experience of Elon Musk's companies violating contracts. The reason given points at the owner rather than at the product or the price.[1]
The size of what OpenAI walked away from is documented. At the start of 2026 Cursor stood among its top five customers by revenue, and by the spring the company estimated the partnership at more than 1 billion dollars in annualized revenue. OpenAI's own reported annualized revenue is more than 40 billion dollars. Set beside each other, the two figures describe a loss that is genuine and that the seller can carry. A firm able to absorb the cost of its principle stands somewhere different from a firm that cannot, and that capacity is what makes the decision available to OpenAI at all.[1]
The party that was not at the table
The people who will feel the change first were not party to it. A developer who chose Cursor built a working day around one editor, its shortcuts, its context handling and the models offered inside it; those models are now decided by an argument between two owners about contracts and trust. OpenAI acknowledged in the same post that ending the partnership may damage its standing with developers, because the decision cuts off a channel many of them use to reach its models. That is an unusual admission: the seller names the group that absorbs the cost and proceeds anyway.[1]
The public argument that followed measured the wrong thing. Cursor founder Michael Truell said OpenAI's models serve about 5 per cent of Cursor user traffic. Thibault Sottiaux, OpenAI's head of core products, answered that token usage is not a proxy for revenue nor value created, and asked Truell to show the arithmetic. Both men were sizing their own company's exposure. Neither figure says what a developer who reaches for an OpenAI model inside Cursor loses on the day it disappears. There is a plain alternative reading and it deserves stating: if Claude and the other remaining models cover the same work, the practical cost to that developer may be close to nothing, and the dispute may concern only the two companies' revenue.[1]
The measurable trace
There is a measurable place where this settles. The model list Cursor publishes for its editor is public, and it changes when access changes. If access to OpenAI's models mattered to the work done inside Cursor, then by the end of 2026 that list either carries a comparable third-party replacement or narrows toward the models its owner controls. Which of the two happens is the trace of who absorbed the cost of this decision, and it needs no statement from either company.[1]
Anthropic's opposite choice runs on the same logic. Cofounder Tom Brown said Claude models would stay available in Cursor, which reads as loyalty to developers until the second figure appears: Anthropic depends on SpaceX for 45 billion dollars of its data centre capacity. A company that has to keep a supplier has no freedom to refuse that supplier's editor. Ownership set OpenAI's decision and ownership set Anthropic's, in opposite directions, and in both cases the developer at the keyboard learned the outcome after it was made.[1]