OpenAI Will Cut Cursor Off After the SpaceX Deal

Suppose you lead engineering at a mid-sized software company and your team of forty lives inside Cursor all day. On Friday morning you learn that one of the four model providers behind the editor is leaving. OpenAI announced on August 28 that it will wind down its contract supplying models to Cursor, with the switch-off set for November 12, 2026, the longest notice the agreement allows. The reason is not technical. It is the editor's new owner.

The deal that triggered it

SpaceX formally closed its $60 billion all-stock acquisition of Cursor on August 14. The purchase grew out of an option SpaceX secured in April: pay roughly $10 billion for a partnership, or buy the whole company for $60 billion later in the year. It chose the latter. By Bloomberg's account this is the largest acquisition of a venture-backed startup ever; Cursor's annualized revenue sits around $2.6 billion. The company now reports into the SpaceXAI team, which puts it under Elon Musk, OpenAI's most persistent adversary.

OpenAI's reasoning, in its own words

The statement is short: "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." Two episodes are cited. In December 2022, shortly after buying Twitter, Musk cut off OpenAI's access to the platform's data. Later, under oath, Musk acknowledged that xAI had violated OpenAI's terms by training Grok on OpenAI outputs (he later called that "partly" true). Thibault Sottiaux of OpenAI reduced it to a sentence: "It boils down to trust."

The contract contained a change-of-control clause allowing termination after an ownership change, and OpenAI is invoking it. The history between the two sides is long: Musk sued OpenAI in 2024 over its shift to a for-profit structure, and a federal jury ruled against him in May 2026 on statute-of-limitations grounds.

What our engineering lead actually loses

Less than the headline suggests. Cursor co-founder Michael Truell said OpenAI models account for only about 5 percent of the editor's AI traffic, and that the company is "speaking with the OpenAI team to resolve this." Cursor users already choose among OpenAI, Anthropic, Google and SpaceXAI models. Anthropic's Tom Brown wasted no time, writing that the company will "continue to expand compute capacity for Claude models in Cursor." So the GPT option disappears from the menu in November; the editor keeps working.

The caveat is that 5 percent of a $2.6 billion business is not a rounding error, and the teams inside that slice are disproportionately likely to be enterprise customers who standardized on GPT models for compliance or evaluation reasons. Those teams have ten weeks to re-benchmark.

The checklist for our engineering lead is short. Look at which model is set as the default in the editor and which model names are pinned in team rules files and agent configurations; if any of them belong to the GPT family, run a handful of real tasks against the Claude or Gemini equivalent before November. If the company holds an enterprise license, ask the account manager whether the contract promises access to any specific model. Switching is an afternoon's work for most teams, but left to the last day it will collide with whatever release is shipping that week.

The lesson for anyone who ships AI features

Cursor survives this comfortably because it never depended on a single provider. If your product or internal tooling calls one model API directly, you are one acquisition, one policy change or one pricing decision away from a forced migration. Abstract the model layer, keep a second provider warm in your test suite, and read the termination clauses in your vendor contract once more with the question "what if someone buys us?" in mind. The cost of that discipline is small. The cost of learning it in November is not.

Sources: OpenAI, The Decoder, Yahoo Finance, TechCrunch