It is tempting to frame the sovereign-AI argument as a fight about who is right about where the value lives. It is more interesting, and more useful, to admit that on the central question we and the biggest platforms are in complete agreement. They have read the board correctly. The model is becoming a commodity. The value is the data and context that people bring to it. We do not dispute a word of that diagnosis. The disagreement — the entire disagreement — is about whose machine the data ends up on.
Read what they do, not what they say
When one of the largest companies on earth open-sources a frontier model, the move only makes sense through this lens. You do not give away your crown jewels. You give away the thing that has become cheap and interchangeable in order to win the thing that hasn't. Open the model, make it free, make it everywhere — and pull the world's data toward the agent you run on your servers, where you can finally own the layer that matters. The open model is the bait. The captured data is the catch. It is a coherent, even elegant strategy, and it rests on precisely the thesis sovereign AI has been arguing: the model is a commodity, the data is the prize.
They give the model away because they, too, know it is not the valuable part. The valuable part is you — and they intend to keep you on their servers. — what the open-source move is really for
So this is not a debate where one side believes in data and the other doesn't. Both sides believe in data. Both sides have organized their entire strategy around owning the context layer. We agreed on the diagnosis before the argument even started. What remains is the only question that was ever really in play, and it is a question about a place: when your agent reaches into your life, where does your life sit while it does? On infrastructure owned by a company whose business is to see it? Or on a machine owned by you, where the company structurally cannot?
The fork is about location, and location is everything
One side's answer: the data pools on the platform's servers. This is sold as convenience, and it genuinely is convenient, and it is also the arrangement in which an always-on agent that knows everything about you runs inside infrastructure whose incentives point at reading you. The agent answers, finally, to the platform. It will be as helpful as is compatible with that, and no more.
The other side's answer: the data stays on hardware you own. Bring whatever model you like — the same commodity model, even the same open weights they published — and let it reach into your context from the outside, under your consent, without taking a copy away. Same open model. Opposite custody. The model is free either way; the only thing that changed is who the agent answers to. On their servers it answers to them. On yours it answers to you. That is the whole fork, and it is decided entirely by location.
Why this framing is the honest one
It would be easier, rhetorically, to cast the big platforms as fools who don't understand their own product. They are not fools. They understand it exactly. That is what makes the disagreement clean instead of muddy: there is no argument to win about the facts, because the facts are shared. There is only a choice to make about the cure. And a choice stated that plainly is harder to wave away than an accusation, because it does not depend on anyone being villainous. It depends only on noticing that “the data is valuable, so we should hold it” and “the data is valuable, so you should hold it” are the same sentence pointed in opposite directions.
Their move even does our work for us. Every time a platform open-sources another model it reconfirms the thesis out loud: the model is not the moat, the data is. We could not buy a better endorsement. The commodity model is their admission and our premise at once. We simply draw the opposite conclusion from the shared fact — that if the data is the thing worth owning, the person it describes should be the one who owns it.
So the line that keeps surviving every rewrite is the honest one: same diagnosis, opposite side. They want your data on their servers. The sovereign answer keeps it on yours. Everything else — the open weights, the free models, the dazzling features — is downstream of that one fork, and the fork is not about who is smarter. It is about whose machine your life lives on.
This is still unfinished, and the far side of the fork has real engineering left to build before “on your machine” is as frictionless as “on theirs.” But the map is clear enough to publish. We don't disagree about the destination's value. We disagree about who should hold the keys to it.