OpenAI Cuts Off Cursor Model Access After SpaceX Deal
OpenAI is winding down the contract that supplies its models to Cursor after SpaceX's acquisition closed, citing Musk's contract history. Shutoff set for Nov 12.
OpenAI is ending the commercial arrangement that supplies its models to Cursor, the AI coding tool now owned by SpaceX. In a notice OpenAI says it delivered to SpaceX on August 28, 2026, the company said it will wind down the contract that gives Cursor direct access to OpenAI models, with a proposed shutoff date of November 12, 2026 — the maximum notice period the agreement allows. The move severs one of the most visible model-supply relationships in the developer-tools market and turns a corporate feud into a concrete operational deadline for millions of engineers.
What OpenAI said
OpenAI framed the decision as a matter of trust rather than performance. In its notice, the company said it “cannot be confident that SpaceX will use our technology within our terms of service,” pointing to what it characterized as a pattern of contract violations by companies controlled by Elon Musk. OpenAI said it is giving the full contractual notice window rather than cutting access abruptly, and that it will not provide future models to Cursor as the relationship winds down.
The practical effect is bounded but real. Cursor routes user requests across models from multiple providers, and OpenAI’s models account for roughly 5% of Cursor’s traffic, according to Cursor chief executive Michael Truell. That figure matters: it means the shutoff is a headache for a slice of Cursor’s users and a symbolic blow to the product, but not an existential threat to a tool that already leans heavily on rival models. Cursor has for some time defaulted much of its heaviest coding work to Anthropic’s Claude family, and the company has been building and fine-tuning its own models on the GPU capacity it now inherits from its parent.
How Cursor ended up owned by a rocket company
The cutoff is the downstream consequence of one of the largest acquisitions in startup history. SpaceX agreed in June 2026 to buy Anysphere, the company behind Cursor, in an all-stock deal valued at roughly $60 billion — the largest acquisition of a venture-backed startup on record. The announcement came just days after SpaceX completed its own record-setting IPO, and it valued Cursor at roughly double the $30 billion the startup had commanded in earlier private rounds. At the time of the deal, Cursor’s annualized revenue was reported near $4 billion, putting the price at about a 15x revenue multiple.
The strategic logic was straightforward: SpaceX wanted an enterprise-software beachhead and a foothold in AI coding to compete with OpenAI and Anthropic, and Cursor gained access to SpaceX’s substantial GPU capacity — the promise of more powerful, cheaper in-house models without renting inference from the very companies it competes with. The deal closed in the third quarter, making Cursor a wholly owned subsidiary and pulling it firmly into Musk’s orbit.
That is precisely the situation OpenAI now cites. Supplying models to Cursor once meant selling inference to a fast-growing but independent customer. After the close, it means supplying a competitor controlled by a figure OpenAI is actively litigating against. The long-running Musk–Altman feud — which has spilled into courtrooms and public accusations on both sides — is no longer background noise to the model-supply business; it is now the stated reason a supply contract is being terminated.
The provider-lock question
For Cursor’s users, the immediate question is what disappears on November 12. Because Cursor is a multi-model client, most workflows can fall back to Anthropic’s or Google’s models, or to Cursor’s own systems, when OpenAI’s are switched off. Developers who explicitly pin OpenAI models for certain tasks will need to migrate those workflows, and Cursor will likely re-route default traffic well before the deadline to avoid a disruptive cliff.
The deeper issue is dependency. The episode is a live demonstration of platform risk in AI tooling: a product can be cut off from a foundational supplier not because of anything it did, but because of who acquired it. That risk runs in both directions. Application builders learn that access to a given frontier model is a commercial privilege that can be revoked when ownership changes; model providers learn that every enterprise customer is a potential future competitor once consolidation accelerates. Cursor’s own hedge — routing across providers and investing in proprietary models — looks less like diversification strategy and more like table stakes after this week.
Not the first friction for Cursor
Cursor has spent 2026 at the center of both the boom and the risk narrative in AI coding. Earlier in the year the tool was implicated in a wave of AI-assisted ransomware attacks tied to the “Aurora” campaign, in which attackers abused agentic coding features to accelerate intrusions — a reminder that the same capabilities that make these tools productive also expand their attack surface. The SpaceX acquisition then reframed Cursor from a neutral, provider-agnostic editor into a strategic asset of one of the most aggressive players in tech.
OpenAI, for its part, has been consolidating its own position across the stack, from buying fleets of Mac hardware to train computer-use agents to expanding its enterprise and consumer footprint. Cutting Cursor off is consistent with a company that increasingly sees the coding-tools layer as territory to win rather than a channel to supply.
What it means
The headline is a contract termination, but the signal is about how the AI market is reorganizing as it consolidates. Three things stand out.
First, model access is becoming a geopolitical-style lever inside the industry. OpenAI is not claiming Cursor misused its models; it is refusing to trust the new owner. As frontier-model providers and application companies increasingly compete head-on, expect more supply relationships to be evaluated less on revenue and more on whether the counterparty is a rival. Neutrality is eroding.
Second, the winner in the near term is Anthropic. Cursor already leans on Claude for much of its heavy coding work, and removing OpenAI as an option concentrates more default traffic on Anthropic’s models and Cursor’s own systems. Every enterprise that watches a peer get cut off has a fresh reason to prize provider optionality — which, in coding specifically, currently favors Claude’s position as the workhorse. OpenAI gives up a small slice of inference revenue to make a point; the point may cost it mindshare among developers who value stability.
Third, platform risk is now a first-order procurement concern. Teams building on any single frontier model should assume access can change with ownership, pricing, or politics, and architect for substitution accordingly — abstracting model calls, testing fallbacks, and avoiding hard dependencies on a single provider’s newest features. Cursor is large enough to absorb the loss of 5% of its traffic; a smaller startup wired directly to one provider might not be.
What to watch next: whether Cursor re-routes default OpenAI traffic well ahead of November 12 and how loudly its users notice; whether SpaceX accelerates its in-house model roadmap to reduce reliance on any external provider; and whether other model suppliers follow OpenAI’s lead in reassessing customers that have been absorbed by direct competitors. The shutoff date is fixed. The precedent it sets is the part still being written.
Keep reading
Chisato · · 6 min read OpenAI Dots: Always-On ChatGPT Agents Explained
OpenAI's Dots are always-on agents with their own cloud computer, powered by GPT-6 Astra. How they work, who gets them, and the safety questions.
Chisato · · 5 min read GPT-6.1 Sol: Pricing, Benchmarks, Access vs Astra
OpenAI's GPT-6.1 Sol launched at DevDay at one-fifth of GPT-6 Astra's token price. The benchmarks, pricing tiers, safety rating, and who can use it.
Chisato · · 6 min read OpenAI Agents API: Codex Harness Now in Public Beta
OpenAI opened its Agents API in public beta on Sept 10, putting the managed Codex harness behind one API call. What it does, how sandboxes work, and pricing.