Listen to the audio version of this article (generated by AI).
OpenAI is walking away from a customer relationship it believed could generate more than $1 billion a year.
That customer is Cursor, an AI coding platform where developers search large codebases, write new features, find bugs, run tests, and hand larger projects to AI agents. Elon Musk’s SpaceX (SPCX) completed its $60 billion acquisition of the company on August 14. Two weeks later, OpenAI announced that it will wind down the contract supplying its models directly to Cursor, with a proposed cutoff date of November 12.
OpenAI didn’t dress it up, either. The company said it simply couldn’t trust SpaceX to play by its rules, pointing to past disputes with Musk’s businesses. The ownership change gave OpenAI a narrow window to cancel – and it took it.
It’s one of the most expensive “no thanks” in recent tech memory.
According to Wired, Cursor was one of OpenAI’s five largest customers at the beginning of 2026. By spring, OpenAI reportedly estimated that the relationship could generate more than $1 billion in annualized revenue.
OpenAI decided the risk of working with Musk outweighed the money.
The feud will grab the headlines. The more revealing question is what OpenAI actually takes with it when it leaves.
Cursor keeps the developers, the workflow, and the ability to route requests to other models.
And that may explain why Musk was willing to pay $60 billion for the company in the first place…
Why OpenAI’s Exit May Not Break Cursor AI
At first glance, OpenAI’s decision looks like a major blow.
Cursor built much of its early success by giving developers one workspace where they could use several leading AI models while they coded. Losing direct access to OpenAI removes one important option from that menu.
But the damage looks a lot smaller than the headline once you dig into things.
Cursor CEO Michael Truell says OpenAI models currently handle roughly 5% of the platform’s user traffic. Now, that figure comes from Cursor itself, and traffic share doesn’t tell the whole story. Some teams have workflows tuned specifically to OpenAI’s models, and switching means retesting or rebuilding parts of their setup.
Even so, Cursor has alternatives. Anthropic responded to the breakup by promising more computing capacity for Claude inside the platform. Cursor also offers models from other providers and increasingly develops its own.
OpenAI is closing the direct pipeline, not every route into Cursor. Developers can still bring their own API access, use the Codex extension, or connect through compatible AI gateways. Everyone else can keep working with Claude, Gemini, Grok, Cursor’s own models, or whatever comes next.
The breakup reveals how the balance of power is changing.
A few years ago, the model provider appeared to own the scarce resource. Applications depended on whichever lab had the strongest intelligence.
Now platforms like Cursor can choose among several capable suppliers.
OpenAI can leave the platform.
Cursor still owns the workflow.
Cursor’s Developer Workflow Is Becoming the Moat
By February, Cursor had reportedly surpassed $2 billion in annualized revenue, double its level three months earlier. Corporate customers accounted for roughly 60% of that total.
The revenue is following the workflow. Developers stay inside Cursor while the models rotate behind the scenes. And the closer Cursor gets to the work, the more leverage it gains over the models supplying the intelligence.
Cursor says it routes hundreds of millions of coding requests each week across different models and providers. Its routing system sizes up each task – what it is, how hard it is, which models have nailed similar work before – and sends it wherever it’s most likely to get done well and cheap.
Routine work can go to a cheaper model. A difficult debugging job can go to a more capable one. Another system may prove strongest at planning or understanding a large codebase.
No model wins every category. When customers use its router, Cursor can decide which one gets the job.
That gives the company something no individual model lab gets to see clearly: a live scoreboard of how every model actually performs on real work.
It sees which outputs developers keep, when they ask for corrections, where one model struggles and another succeeds. And it can use those signals, within users’ privacy and retention settings, to improve how future work gets routed.
This is what Musk bought.
Cursor owns the interface, the enterprise relationship, and the decision about which model gets each job.
The AI Model War Is Becoming a Distribution and Routing War
OpenAI’s exit exposes the new balance of power.
Model labs still compete by building better intelligence. Platforms like Cursor increasingly influence where that intelligence gets used and which provider receives the work.
A clearly superior model can still command premium prices and pull in developers. But intelligence only makes money when it reaches users inside a real workflow.
Cursor already owns that workflow.
The best model for a particular task may come from OpenAI today, Anthropic next quarter, and a new open system a year from now. Cursor can make those changes largely invisible by routing work behind the scenes.
The user keeps working in the same place while the supplier changes underneath.
Cursor’s own research shows where this could lead. The company says its router can send simpler work to lower-cost models and reserve premium systems for tasks where their added ability justifies the price. In company-run A/B tests across millions of requests, Cursor says its router delivered frontier-quality performance at 60% lower cost.
That is a very different AI market from the winner-take-all model war many imagined.
One model handles routine coding. Another plans a complex software migration. A third debugs visual interfaces. A fourth runs sensitive work inside a private environment.
The platform decides who gets called into the game.
That makes control of the workflow a powerful bargaining chip. Model companies need distribution to turn intelligence into revenue – and the platforms own the distribution.
OpenAI walked away from Cursor because it did not trust the new owner.
Anthropic immediately leaned in.
That is what competition looks like when the customer owns the valuable real estate.
Why SpaceX Paid $60 Billion for Cursor
SpaceX paid $60 billion to own the place where developers build software.
That gives Musk a direct outlet for the enormous amount of computing power SpaceXAI is assembling. xAI can train Grok. SpaceX can build the infrastructure. Cursor gives that intelligence a direct route into a large group of paying users.
In the tech economy, developers punch far above their weight. They build the software used by banks, retailers, factories, hospitals, governments, and consumers. The models and tools they choose today can shape which clouds, chips, and platforms receive traffic tomorrow.
Cursor made that logic clear when the acquisition closed. The company said SpaceX would give it access to what it called the world’s largest GPU fleet, allowing Cursor to build stronger models at lower cost.
Then it summed up the deal in one sentence: “Cursor will be one place where that intelligence becomes useful.”
Cursor gives Musk control of the workflow. OpenAI’s exit reveals the limit of that control – and another important part of the investment thesis.
A model supplier can leave. Cursor can redirect much of the work toward Anthropic, Google, Grok, its own systems, or whatever capable model comes next. The switch may carry costs, and certain workflows may need to be rebuilt. But the exits are clearly marked.
The physical bottlenecks underneath Musk’s companies are much harder to route around.
The Partnerships Musk Cannot Replace
SpaceX cannot replace an advanced chip foundry with a software update. Tesla cannot swap out every power semiconductor, optical component, sensor, manufacturing tool, and specialty material whenever a supplier walks away.
Those capabilities require factories, scarce technical expertise, years of qualification, and enormous amounts of capital. Musk can pull more production in-house. But he can’t conjure an entire supply chain on command.
I’ve spent months mapping the suppliers that SpaceX, Tesla (TSLA), xAI, and Musk’s other companies still rely on.
This evening, at 8 p.m. Eastern, I’ll walk through that system during a free InvestorPlace workshop. We’ll dig into where Musk remains dependent on outside partners, where the hardest bottlenecks are forming, and which suppliers could feel the biggest impact as his spending accelerates.
OpenAI just showed us the layer Musk can route around. Tonight, we’ll show you the suppliers he still has to pay.
