The American technology industry has now formally discovered that AI is not a cellphone plan.
NVIDIA, IBM, Meta, Microsoft, Dell, Palantir, Box, Hugging Face, Mistral and many others have just signed an open letter on open-weight AI.
Their argument is that organisations must be able to download, inspect, modify and run AI models on infrastructure they control.
They explicitly connect open weights with control over data, freedom from vendor lock-in, specialized internal capabilities and the sovereign use of AI across the economy.
They also argue that companies should match the right model to the right job at the right cost, reserving frontier-scale systems for genuinely frontier problems.
This is almost exactly the argument I made in June in AI Is Not a Cellphone Plan.
AI is not a service you simply subscribe to and forget; it becomes part of your technical and organisational capability.
The interesting part is not that open-model advocates are saying this, but that cloud, hardware and enterprise-software giants are now saying it together.
The letter is written around American AI leadership, but the argument is even more important for European companies that cannot outsource control of critical systems indefinitely.
The industry may still sell AI like a monthly plan, but it has now put in writing that lasting value comes from owning the choices, knowledge and capability around it.