Let’s talk about the software team that lost its engine in November because of a deal that was signed in June by people they’ve never met.
On the twenty-eighth of August, OpenAI told SpaceX it was winding down the contract that puts OpenAI models inside Cursor. The shutoff date is the twelfth of November. That’s nine days from now.
Read the letter and it’s polite. Maximum notice allowed by the contract. Deep respect for the team. Ready to support developers through the transition.
Then read the second paragraph and you find the mechanism. There was a change of control. Cursor got bought. The contract had a limited window to cancel after that, and OpenAI is using it.
SpaceX agreed to buy Anysphere, the company behind Cursor, in June. Sixty billion dollars, all stock. It closed in the third quarter and the division now sits inside something called SpaceXAI.
So the supply of the thing Cursor sells was decided by a share transaction, not by anything Cursor did. And OpenAI went one step further and said it won’t give Cursor its next model, Astra, at all.
Now, picture the standup.
Six developers, Monday morning, a client demo booked for Thursday. For two years they’ve tuned everything around one model: the review prompts, the onboarding docs, the little internal benchmark they run before every release. Nobody in that room negotiated anything. They picked the tool that felt best and got on with the work. And this week the tool is still there, and the engine underneath it is not the engine they chose.
That’s the scene. Here’s the pattern.
For as long as software has had platforms, the switch has existed. Your app gets pulled from the store. Your ad account gets banned. Your API version gets deprecated. We’ve learned to live with it, mostly because it was always framed as your fault. You broke a rule. You didn’t read the policy. You were on a version that died.
This one is different.
Nobody at Cursor broke a rule. The clause that took the engine away was triggered by who owns them. The supplier isn’t judging your behaviour. It’s reading your shareholder register.
And every one of those clauses was signed back when you were small, by a person who skimmed it, because the deal needed to close and the lawyers were busy with the big one.
Three pieces of this are worth pulling apart.
First, the clause itself. Change of control, termination for convenience, suspension on suspicion, model availability “at our discretion.” These sentences sit in supplier agreements everywhere now, and almost nobody can tell you what’s in theirs. Not because they’re careless. Because the contract arrived as a ninety-page PDF at eleven at night and the only question anyone asked was the price.
Second, the frontier. You never bought a model. You bought access to the road that leads to the next one. The version you’re running today is not the product. The product is being first in line when something better lands. So when a supplier says you don’t get Astra, they haven’t degraded your service next month. They’ve quietly removed you from the future, and left the old thing running so the graphs don’t dip.
Third, the notice. Eleven weeks sounds generous. In a contract, it is generous. In a product cycle, it’s nothing. You can’t re-architect an onboarding flow, retrain a team, re-benchmark a workflow and renegotiate with clients in eleven weeks. You can only scramble, and pay for the scramble.
Now here’s the part most commentary will skip.
None of this means OpenAI is wrong. They gave a reason — that they can’t be confident the new owner will stay inside their terms, based on how Musk’s other companies have behaved. They may be entirely right on the facts. That’s not the point.
The point is that the customer had no vote, no seat, and no warning until the letter arrived. A board decision two layers above them re-specified what they sell. And they found out at the same time as everybody on the internet.
If you run a business on this stuff — and most of you do now, whether it’s in the product or just the back office — this is your exposure, and it’s bigger than the model’s price.
Your AI isn’t a purchase. It’s a supply relationship with a kill date you didn’t negotiate.
And it’s already being priced. If you ever sell your company, one of the first questions in diligence will be: what happens to your engine if your vendor gets acquired, or decides you’re now a competitor? You want an answer that isn’t a shrug. The answer itself is worth money. The shrug is a discount.
So what do you actually do about it this week?
Pull the contract and read four things: change of control, termination, suspension, and anything about model or version availability. Not the whole document. Those four. Write the dates and the trigger conditions into a calendar with a name next to them, because a clause nobody is watching is a clause that fires on a Tuesday.
Then build a second path. Even a bad one. Route twenty real tasks through another provider this month and look at the output honestly. An untested fallback is not a fallback, it’s a rumour you tell the board.
Keep your own test set. A small pile of jobs you know the right answer to, run against whatever you’re using. Then a model swap becomes a measurement instead of a vibe. That one habit is the difference between a supplier change being an incident and being a Tuesday.
And say it out loud to clients. Where the engine comes from, what happens if it changes. I know it feels like confessing to a weakness. It isn’t. In a market where everyone else is pretending the ground is solid, the person who names the fault line is the one people trust with the building.
One more thing, and it’s the quiet version.
The loud failure is a letter with a date. The quiet failure is worse. Your product starts behaving like your supplier’s priorities, one small default at a time, and no letter ever arrives. You just wake up one quarter with a tool that does what the vendor wants and not what your customer asked for, and nobody can point to the day it changed.
You can’t stop a board meeting in another country. You can make sure that meeting isn’t the only source of the thing you sell.
You’ve got this.