Cursor Loses Access to OpenAI's Models After the SpaceX Takeover — the Lesson About AI Vendor Dependency

OpenAI is invoking a change-of-control clause

Cursor, one of the most widely used AI coding apps, operated independently, with access to models from multiple vendors — OpenAI, Anthropic, Google. In April 2026, SpaceX and Cursor signed an option agreement giving SpaceX the right, but not the obligation, to acquire parent company Anysphere at an implied equity value of $60 billion in stock — plus a separate agreement under which SpaceX would supply Cursor with compute capacity. On June 16, days after SpaceX went public, SpaceX exercised the option and the parties signed the merger agreement. The transaction closed on August 14.

On August 28, OpenAI notified SpaceX that it intends to wind down the contract supplying models to Cursor, with a proposed shutoff date of November 12 — the maximum notice period the contract allows. Importantly, OpenAI states explicitly, in its own post, that this is not about a breach committed by Cursor. The company is invoking a standard "change of control" clause — a contractual mechanism that gives one party the right to reconsider a relationship when the other party's ownership changes, regardless of that party's own behavior.

OpenAI's stated reason: "we cannot be confident that SpaceX will use our technology within our terms of service," based on prior experience with Elon Musk's companies. OpenAI explicitly cites two precedents: Twitter/X breached the terms of OpenAI's contract after Musk's takeover, and Musk admitted under oath, earlier in 2026, that xAI — now part of SpaceX — had used distilled OpenAI data for training, in violation of those same terms.

Not the first time — and that matters more than the Musk–Altman feud

If this were an isolated incident, it would just be a story about the Musk-Altman rivalry. It isn't. In June 2025, Anthropic sharply cut Windsurf's — another popular AI coding tool's — direct access to the Claude 3.7 Sonnet and Claude 3.5 Sonnet models, with less than a week's notice. The reason: reports, via Bloomberg, that OpenAI was about to buy Windsurf for $3 billion. Jared Kaplan, Anthropic's co-founder and chief science officer, said at the time: "it would be odd for us to sell Claude to OpenAI" — and that the company preferred to reserve its compute for "lasting partnerships."

The mechanism differs: with Windsurf, Anthropic acted at its own discretion, based on rumors, before any deal had officially closed. With Cursor, OpenAI is invoking an explicit contractual clause, after the acquisition was already final. But the structural risk exposed is the same in both cases: an intermediary's access to an AI model can be cut because of who becomes its owner, not because of its own behavior. Two public instances in just over a year are no longer an isolated incident — they're starting to look like a market pattern.

Cursor shows how to reduce vendor risk — but it doesn't walk away unscathed

Cursor wasn't built on a single vendor to begin with. The app offers access to Claude, GPT, Gemini and other models — a multi-model architecture that is now proving to be exactly the kind of resilience it needs. On August 29, Cursor CEO Michael Truell responded publicly on X: OpenAI's models account for roughly 5% of Cursor's user traffic, and the company is talking with the OpenAI team to resolve the situation.

That figure deserves a caveat, not unconditional acceptance. An OpenAI engineer, Thibault Sottiaux, publicly challenged how representative it is: tokens aren't a direct proxy for revenue or value created, and OpenAI's models are among the most token-efficient per task — so a percentage calculated on traffic could understate how much OpenAI's models actually mattered in users' workflows. Neither side has published the underlying math.

Anthropic wasted no time. Tom Brown, Anthropic's co-founder and chief operating officer, said Cursor has been a trusted partner of the company since Claude 3.5, and that Anthropic will "continue to increase compute" allocated to Claude in Cursor. The right takeaway isn't that Cursor "got off easy" — it's that Cursor is far better positioned than a company built 90% on a single model vendor would be. Diversification didn't prevent the conflict, but it sharply limited its impact.

What to check before you build a critical process on top of an AI model

The term "vendor lock-in" usually describes the risk that it's hard for you to leave a vendor. The Cursor case shows the less-discussed opposite risk: the vendor can leave you, through a contractual decision you don't control and that has nothing to do with anything you did. For any company running a business process on top of an AI model — directly or through an intermediary tool — four concrete checks are worth doing before that process becomes critical, not after.

What termination rights does the model vendor hold, and what exactly triggers them — change of control, competitor clauses, non-compliant use. Are there "change of control" clauses that could change your access if the intermediary vendor — not the model itself — changes ownership. Can the workload be moved to another model, if needed, without rebuilding the application from scratch. And, most important: how much of your process depends on capabilities specific to a single vendor, not replicable elsewhere.

For companies in Romania, the risk is already on the radar — in the 2026 Allianz Risk Barometer, AI-related risk ranks 2nd among business risks, right behind regulatory and legislative change; globally, AI-related risk climbed from 10th to 2nd place. Vendor dependency is one facet of that risk. For companies, multi-model is no longer just a cost-and-performance strategy. It's becoming an operational-continuity strategy. We detail what a model-vendor-independent execution architecture looks like in practice in our guide to operational AI agents.

Sources: ↗ OpenAI — our decision on Cursor · ↗ TechCrunch — the SpaceX-Cursor deal · ↗ CNBC — the access cutoff · ↗ TechCrunch — the Windsurf precedent