OpenAI is terminating Cursor's direct access to its AI models effective November 12, 2026, after SpaceX completed a $60 billion acquisition of the coding startup. OpenAI cites distrust of Elon Musk-linked entities honoring contract terms and is forgoing over $1 billion in projected annual revenue; Cursor says the impact is limited since OpenAI models drive only about 5% of its traffic.

SpaceX completed its $60 billion all-stock acquisition of Cursor's parent Anysphere on August 14, 2026 ^[1]. Two weeks later, on August 28, OpenAI notified SpaceX it intended to wind down the model-supply contract, invoking a change-of-control clause that gave it a limited window to cancel following the ownership change ^[2]. Under the terminated deal, Cursor's direct access to OpenAI's models - a relationship dating back nearly four years - will fully end November 12, 2026 ^[3][4], with Cursor among OpenAI's earliest users ^[2]. OpenAI went further than a simple wind-down: its next model, Astra, will not be made available to Cursor at all ^[5]. The acquisition itself followed a structure agreed months earlier: SpaceX held an option to either buy Cursor outright for $60 billion or pay $10 billion for a compute-partnership arrangement, and had already given Cursor access to its Colossus supercomputer, roughly 1 million H100-equivalents, to help train Cursor's own Composer model before the deal closed.

OpenAI's official framing is narrow: it cannot be confident SpaceX will use its technology within the terms of service, pointing to a history of Musk-linked companies, including Twitter and xAI, allegedly violating contract terms ^[3][5]. The specific evidence OpenAI leans on is that Musk admitted under oath that xAI used OpenAI model outputs to train Grok via distillation ^[6]. That detail reframes the dispute: the real anxiety isn't hypothetical future misuse so much as a concrete precedent of a Musk company allegedly mining a rival's model outputs to build a competitor. Commentary following the announcement split along similar lines - much of it treated the 'trust' language as a thin cover for fear that Cursor's accept/reject coding signal, now sitting inside the same corporate structure as xAI, could feed Grok's training pipeline, while a smaller contrarian view pushed back that framing a paying customer's use of purchased inference as a 'distillation attack' stretches the concept past its breaking point. Either way, the decision reads less like routine vendor risk management and more like OpenAI drawing a hard line around its most defensible moat as capability gaps between labs narrow.

This is not the first time a foundation model lab has yanked access from a coding tool mid-acquisition. In June 2025, Anthropic cut off Windsurf's first-party Claude access with less than a week's notice while OpenAI was in the process of acquiring Windsurf for roughly $3 billion ^[7]; that deal ultimately collapsed, and Windsurf's remaining team and technology were absorbed by Cognition, maker of Devin ^[8][9]. Now the roles are inverted: Anthropic is publicly positioning itself as Cursor's stable partner, with co-founder Tom Brown pledging to increase Claude compute and raise usage limits as OpenAI exits ^[2]. The pattern that emerges across both episodes is the same - when a rival lab acquires a popular coding tool, the incumbent model supplier treats continued access as a strategic liability rather than a revenue stream worth protecting, regardless of which lab is playing which role.

By Cursor's own account, the practical disruption is modest: OpenAI models serve roughly 5% of Cursor's user traffic, with the bulk already running on Anthropic's Claude and Google's models, and CEO Michael Truell says the two companies are still talking to find a resolution ^[2]. That said, YouTube commentary on the deal has argued the raw percentage may understate GPT's real importance, since complex coding tasks are reportedly still routed to GPT while cheap bulk work goes to Cursor's own Composer model or Grok. Yet OpenAI is nonetheless forgoing a partnership it internally projected at over $1 billion in annual revenue ^[2][10], and the move has been widely read as another chapter in the long-running Altman-Musk feud rather than a purely commercial calculation ^[11]. The gap between the traffic figure and the price OpenAI is willing to pay for its trust argument is the real story: it signals that AI labs are starting to treat model access itself as a lever to fight rivals through customers, not just through capability races, which is precisely the dynamic the 2025 Windsurf episode foreshadowed ^[7].