OpenAI Is Cutting Cursor Off on November 12. The Number That Matters Is 5%, Not the $60B Price Tag.
OpenAI Is Cutting Cursor Off on November 12. The Number That Matters Is 5%, Not the $60B Price Tag.
On August 29, OpenAI told Cursor it's ending direct API access to its models on November 12. The reason isn't a security incident or a technical dispute. It's that SpaceX closed its roughly $60 billion all-stock acquisition of Anysphere, Cursor's parent company, earlier this month, and OpenAI says it can no longer trust a Musk-controlled company to honor its terms of service. Cursor's CEO responded within hours by pointing out that OpenAI's models handle about 5% of Cursor's traffic today. That 5% is the real story here, not the acquisition price or the feud.
What actually happened and why
OpenAI's stated reason is contractual trust, not a product complaint about Cursor itself. In its own announcement, the company said it "cannot be confident that SpaceX will use our technology within our terms of service," and pointed to what it describes as a pattern of Musk-affiliated companies not honoring contract terms. This is a platform vendor pulling API access over who owns the downstream company, which is a different kind of risk than an outage or a price hike. You can plan around a price hike. You can't easily plan around "the company you buy access from decides they don't like your new owner."
Anthropic moved the same day. Tom Brown, Anthropic's co-founder, posted publicly that Cursor "has been a trusted partner of Anthropic since Sonnet 3.5" and that Anthropic will "continue to increase compute to support Claude models in Cursor." Anthropic also announced a new compute partnership with SpaceX, separate from the Cursor dispute, that includes interest in developing orbital AI compute capacity together. So the same week OpenAI is walking away from a customer relationship, Anthropic is deepening one with the same acquirer. That's not a coincidence, it's competitive positioning happening in real time.
The 5% number is doing all the work
Cursor CEO Michael Truell's response downplayed the cutoff, and the number he gave explains why he could. OpenAI models reportedly handle around 5% of Cursor's model traffic now, a share that's shrunk substantially from when GPT models were Cursor's default backend. Ramp's corporate spend data backs up the direction of that shift: Cursor's own share of the AI coding spend category has been sliding against Claude Code's rise over the past year, even before this cutoff.
That's the part worth sitting with if you're a Cursor user, or if you route any part of your own product through a single AI API. Cursor could absorb a full vendor cutoff from what used to be its primary backend because it had already diversified traffic away from that vendor for competitive and pricing reasons unrelated to this dispute. If OpenAI had done this eighteen months ago, when GPT models were still the default, it would have been existential for Cursor rather than a Tuesday headline.
The dependency risk nobody budgets for
Most solo operators think about vendor risk in terms of uptime and pricing. Will the API go down. Will the per-token cost double. Multi-model routing setups (OpenRouter, LiteLLM, a thin internal abstraction layer) get built to hedge exactly those two things. What almost nobody budgets for is a vendor revoking access over a corporate ownership change that has nothing to do with your usage, your payment history, or your product.
This is worth naming precisely because it's not hypothetical anymore. It just happened to a company with a $60 billion valuation and enterprise-grade contracts. If OpenAI, Anthropic, Google, or any other model provider decided tomorrow that a company you depend on (your hosting provider, your auth provider, your payments processor) had an ownership change they didn't like, the same mechanism applies. The protection isn't a better contract. It's knowing, concretely, what percentage of your actual traffic or revenue any single vendor carries, the same number Truell had ready to quote within hours of getting the notice.
What I'd actually do
Run the exercise Cursor's CEO clearly already had an answer for: what percentage of your product's actual function depends on any one API you don't control. Not "which providers am I integrated with," the real number, measured in requests or revenue, not in how many providers appear in your config file. If one vendor is carrying more than half your traffic and you have no fallback that's been tested in the last quarter, that's the actual risk, not a slow API response.
The honest complication is that this isn't free. Building and maintaining a real fallback path costs engineering time you could spend on features, and for a lot of solo products the actual probability of a Cursor-style cutoff is low enough that the insurance isn't worth the premium. I don't multi-route everything I build either. But I do know, for the handful of things where a single API going away would actually break the product, what the fallback is and that it's been tested this year, not just written down. That's the bar. If you can't answer "what's my 5% number" for your own stack right now, that's the thing to fix this week, not the acquisition drama.
Author
Lukas
@lukcombinator