SDSignal Desk

OpenAI Cut Off a Billion-Dollar Customer to Avoid Elon Musk

Sep 3, 2026, 9:42 AM · WIRED

Image: WIRED

OpenAI estimated Cursor at more than $1 billion a year in revenue and still killed the partnership after SpaceX bought the editor. That is a lawsuit hanging over a product decision.

Why it matters

Maxwell Zeff reports that at the start of 2026 Cursor was one of OpenAI's top five customers by revenue, and that by spring OpenAI internally estimated the partnership at more than $1 billion in annualized revenue. Those figures had not been previously reported. Last Friday OpenAI said it would wind down the partnership because it cannot be confident SpaceX will stay inside its terms of service, citing 'experience with Elon Musk's companies violating contracts.'

SpaceX acquired Cursor in a $60 billion deal. Cursor founder Michael Truell, now leading teams at SpaceX, posted that OpenAI's models serve about 5 percent of Cursor user traffic. OpenAI's Thibault Sottiaux replied that token usage is not a proxy for revenue or value. Anthropic cofounder Tom Brown said Claude will stay in Cursor.

The Signal Desk read

Walking away from a billion-dollar run-rate customer is not a terms-of-service footnote. It is OpenAI deciding that Musk-as-counterparty is a bigger risk than the revenue, the developer channel, and the optics of cutting off a tool a lot of its users actually live in. The public reason is contract violation history and an apparent Musk deposition remark that xAI used OpenAI models to train its own. The private reason, which Zeff puts in plain sight, is that Codex and Cursor are already competing for the same seats, and a $60 billion SpaceX owner makes the old coopetition arrangement look naive.

Signal Desk's read: the $1 billion estimate is the number that will travel, and it should. It tells you OpenAI would rather take a visible hit on the eve of a 2027 IPO than keep a pipe into a Musk-controlled product that could distill weights. Distillation risk is the actual fear. Terms of service is the language you use when you cannot say that in a customer email.

Truell's 5 percent line is a counter-narrative, not a rebuttal. If OpenAI models are a small slice of traffic they can still be a large slice of the bill, which is exactly what Sottiaux was pointing at. Anthropic staying in Cursor is the more revealing move. Last year Anthropic cut off Windsurf when OpenAI was rumored to be buying it. Now Claude stays, and Zeff notes Anthropic depends on SpaceX for $45 billion of data center capacity. That is not philosophy. That is a landlord you do not evict.

OpenAI's own startup fund was in Cursor's 2023 seed. It later talked acquisition, never seriously. Truell appeared in GPT-5 launch marketing. This is a relationship that went from investment to partnership to competitor to cutoff in three years, and Musk was the event that forced the last step, not the first.

Context

OpenAI is preparing to go public next year and Zeff cites more than $40 billion in annualized revenue across subscriptions, ChatGPT ads, and Codex. Losing Cursor hurts less than it would have two years ago, which is probably why the company felt it could do this in public.

Who feels it

Developers on Cursor
OpenAI models will wind down there. Claude is the remaining first-party path. Expect a messy few weeks of default-model churn.
OpenAI investors
A $1 billion customer walked because of counterparty risk. That is a governance story as much as a revenue story.
Anthropic
Staying in Cursor after cutting off Windsurf makes the SpaceX compute relationship the real constraint. Watch whether that lasts past the next round of contract talks.

What to watch

  1. The actual wind-down date for OpenAI models inside Cursor, and what replaces them as the default.
  2. Whether SpaceX or xAI ships a first-party coding model into Cursor within a quarter.
  3. Any disclosure, in IPO paperwork or otherwise, of how concentrated OpenAI's top-five customer revenue still is.

Read the original

Continue at the source.

WIRED

Companies: OpenAI