Airbus just made it a procurement criterion.
On May 28, the company signed a contract with Mistral AI that lets the model weights and the training data stay on its own servers.
That clause is missing from almost every AI vendor contract a US or UK plant signed this quarter.
Before we get into what changed, quick question:
How does your plant decide where AI vendor data is processed?
👋🏻 I'm Leonardo Ubbiali. This week we're looking at what Airbus picked Mistral for, and why the same clause is missing from your AI vendor contracts.
Airbus and Mistral AI signed in Paris on May 28, at Mistral's AI Summit
The contract gives Airbus the full Mistral product suite, and Airbus picks where every model runs. On its own servers, on a trusted cloud, or anywhere else it wants.
Catherine Jestin, EVP Digital at Airbus, said the deal builds the foundation for trusted and responsible AI across Airbus's products and services.
The use cases run across technical document automation, engineering simulation, and onboard object recognition for commercial aircraft, helicopters, defense, and space.
That clause is what made Airbus pick Mistral.
What's In The Contract
The contract spells out four things most AI vendor contracts do not:
Where the data is processed.
Who owns the model weights once Mistral has trained on Airbus engineering data.
Which country's law governs the deal.
What happens to those weights the day the contract ends.
None of that is standard in the AI vendor template most plants are signing right now.
And Airbus is not alone. BMW signed the same kind of deal with Mistral in May to train models on its crash simulation archive.
Stellantis started its Mistral partnership in 2024 and expanded it at Italian Tech Week last October.
ASML wrote a €1.3 billion check into Mistral's Series C last September and now owns roughly 11% of the company.
Four of Europe's biggest industrial buyers picked the same European AI vendor in twelve months.
I have looked at the AI vendor template most US and UK plants are signing this year.
The standard terms hand over the same 4 things Airbus negotiated for: where the data is processed, ownership of the fine-tuned model, the choice of governing jurisdiction, and what happens to the weights when the contract ends.
Most plants sign it without asking.
Airbus, BMW, ASML, and Stellantis refused that default.
They put a new procurement clause in front of every AI vendor that wants their business.
The plants getting this right are doing the same four things this quarter.
They are adding the data residency clause to every RFP. The engineering files already sitting in foreign hyperscaler clouds are getting mapped this quarter.
On-premise and sovereign AI vendors are going on the qualified list before the next renewal.
Model-weight ownership is going into every new contract, because the data that trains the model is the asset you cannot get back.
Anyone signing a multi-year AI contract this quarter without those four clauses is locking in the wrong answer.
Five Things You Can Do This Quarter
The problem: You need to know whether your AI vendor contracts protect the engineering data and operational data flowing through them.
What you need: A list of every AI tool, copilot, or agent your plant uses, where each one's data is processed, who owns the model weights, and which jurisdiction's laws apply.
The Prompt (copy this):
I'm a [YOUR ROLE] at a [FACILITY TYPE] manufacturer. The following AI tools are in use at our plant: [LIST EACH TOOL, ITS VENDOR, AND ITS PURPOSE]. For each one, the data is processed in [LIST JURISDICTION OR CLOUD] and the model is hosted [LIST WHERE].
Tell me: Which contracts have data residency clauses I should add or strengthen? Which vendors give me model ownership rights after I fine-tune on plant data? Which jurisdictions' laws apply to my training data? What clauses should I require in the next renewal to put model weights and training data under my control?
What you'll see:
A vendor-by-vendor sovereignty audit, a list of contract terms to add, a jurisdiction map of where your data currently sleeps, and a renewal checklist.
NTT DATA 2026 Global AI Report: A Playbook for Private and Sovereign AI
The 95 / 29 / 60 stats give you the size of the sovereignty gap. The playbook section walks through what early movers are doing differently in their AI architecture.
Time to value: 25 minutes
Airbus picked where its model weights would live before it picked Mistral. Where will the weights of your next AI tool sit?
Hit reply. I read every email.
Leo