On 15 August SpaceX closed a 60 billion dollar all stock acquisition of Cursor. Four days later Stripe confirmed it had bought OpenRouter. Both companies sit inside enterprise AI stacks, and in neither case did the customers get a vote. Your AI vendor contract has a change of control position whether or not anyone negotiated one.
Five Days, Two New Owners
SpaceX completed its purchase of Cursor, the code editor that writes and completes software alongside developers, on 15 August, paying in SpaceX stock. On 19 August Stripe confirmed it had acquired OpenRouter, the platform corporate customers use to direct AI workloads across roughly 400 models from Anthropic, Google, OpenAI and others, and to manage what those calls cost. Stripe declined to comment on terms. The New York Times reported a figure above seven billion dollars.
Neither is a distressed sale or a cautionary tale. Both are healthy companies being bought at scale, which is the point. The vendor you diligenced for financial stability can be acquired precisely because it is doing well, and the acquisition changes things that stability never threatened.
We looked at the survival question a week ago in our piece on what AI compute financing does to vendor risk. This is the other failure mode. Not the vendor going away, but the vendor staying and answering to somebody else.
What an Acquisition Changes That a Durability Check Does Not Catch
A durability assessment asks whether a vendor can keep serving you. It is the right question and it does not cover any of the following, all of which move on the day a deal closes.
The roadmap. Priorities get re-set against the acquirer strategy. A feature on next quarter plan because you asked for it competes against integration work nobody outside the new parent can see.
The price. Nothing forces a new owner to honor a rate card that was set to win market share, once market share is no longer the goal. Renewal is where this shows up, which is usually the first moment you can push back and the last moment you want to be discovering the problem.
Who touches your data. An acquirer brings its own infrastructure, its own security posture and its own list of sub-processors, meaning the third parties permitted to handle your data on the vendor behalf. That list can change without the contract changing, if the contract never constrained it.
The people. Founders sign retention packages. Support engineers do not. The institutional knowledge of your specific deployment is held by individuals whose reasons for staying just changed.
The AI Vendor Change of Control Terms That Decide What Survives
These belong in the agreement you are negotiating now, not in the amendment you will try to negotiate from a weak position later. None of them are exotic and all of them are ordinary in mature software contracts, which is the argument for asking.
Assignment and change of control
Most agreements let a vendor assign the contract to a successor freely. Change of control means an acquisition, merger or sale of substantially all assets, and an assignment clause that is silent about it hands your agreement to whoever buys the company. Ask for notice within a defined number of days, and ask for what you actually want next: either consent rights, or a termination right that the change itself triggers.
Price held through the transition
A rate that survives the close for a defined period, usually the remainder of the term plus one renewal, converts an acquisition from a budget event into a scheduling one. This is the term vendors concede most readily, because at signing they do not expect to be acquired either.
Data commitments that bind the successor
Say plainly that the data protection terms, the sub-processor list and the deletion obligations survive assignment and bind the acquirer. Add notice before any new sub-processor is added, with a right to object. Without it, the answer to where your data now lives is whatever the new parent decides.
An exit that is actually usable
A termination right is worth what your ability to leave is worth. That means export in a documented, machine readable format rather than a support ticket, a defined transition assistance period, and clarity on what happens to anything the vendor generated for you. A right to walk out of a building with no doors is not a right.
The Contracts You Have Already Signed
You cannot retroactively add terms, so the exercise is different: find out what you agreed to. Pull the assignment clause on every AI vendor above a materiality threshold you set, and sort them into ones that require your consent, ones that require notice, and ones that are silent. The silent ones are your exposure, and the fix is renewal. Put the change of control terms on the renewal checklist for each, and price the switching cost now while nothing is urgent, because the number itself will tell you which relationships are worth hardening and which were never that hard to replace.
Key Takeaways
- Two AI companies embedded in enterprise stacks changed owners inside five days in August 2026, and customers had no vote in either.
- A vendor durability assessment asks whether a vendor survives. It does not cover what changes when a healthy vendor is acquired: roadmap, price, sub-processors and staff.
- Four terms decide what survives a close: assignment and change of control, price held through the transition, data commitments that bind the successor, and an exit that is usable in practice.
- These are ordinary terms in mature software contracts, and vendors concede them most easily at signing, before either side expects an acquisition.
- For contracts already signed, audit the assignment clauses, treat the silent ones as the exposure, and move the terms onto the renewal checklist.
Frequently Asked Questions
Is a termination right on change of control realistic to get?
More often than teams expect, particularly below the largest vendors, and most often when it is asked for at signing rather than at renewal. Where a hard termination right is refused, the fallback that usually clears is notice plus a price hold plus an extended transition assistance period. That combination gets you most of the protection, because what you generally want is not to leave immediately but to leave on your schedule rather than on the acquirer schedule.
Does this apply to the large model providers too?
The terms matter there, but the balance of power is different and the standard agreement is rarely negotiable below a significant commitment. For those relationships the practical protection is architectural rather than contractual: keep the integration portable enough that switching is a real option, which is what makes any of these terms enforceable in the first place. The contract terms do most of their work with the layer above the model, where the vendors are smaller and the switching cost is quietly higher.
How do we set the materiality threshold for the audit?
Not by spend. Sort by what breaks if the vendor changes behaviour: whether it sits in a customer facing path, whether it holds personal data, and how long a replacement would take. A small annual contract for something wired into a production workflow deserves more scrutiny than a large one for a tool that a team could stop using on a Friday.
Sources
- TechCrunch, "SpaceX officially closes its Cursor acquisition," 2026. Link.
- Payments Dive, "Stripe, OpenRouter finally strike a deal," 2026. Link.
- Fortune, "Stripe clinches over $7 billion deal to buy AI firm OpenRouter," 2026. Link.
Next Steps
If nobody on your team can say what happens to your AI vendor agreements when the vendor is bought, the answer is currently being written by other people. Stable Solutions reviews AI vendor agreements for change of control exposure and builds the integration portability that makes an exit right worth having. Explore our Digital Growth Strategies or contact our team to audit the agreements you have already signed.