Watch where the money is pointed and you will usually find where people think the value lives. Right now almost all of it is pointed at the model — the weights, the parameter count, the benchmark leaderboard, the next hundred-billion-dollar training run. The assumption underneath the spending is that whoever builds the smartest model wins. That assumption is already wrong, and the people spending the money know it. The tell is that some of them are giving their models away.
When a company open-sources a frontier-class model, it is not being generous. It is telling you, with its actions, that the model is not where the value is. You do not hand out the thing you are trying to sell. You hand out the thing that has become a commodity — cheap, interchangeable, a means to an end — in order to pull people toward the thing you actually intend to own. And the thing they intend to own is not on the leaderboard. It is you.
Capability is converging; context is not
Capability is converging. Each new frontier model is a smaller leap over the last, the open models trail the closed ones by months instead of years, and a model good enough to do most of the work you need is already cheap enough to run. This is what commoditization looks like: the product stays impressive while it stops being scarce. Electricity is miraculous and nobody chooses a provider on the brilliance of their electrons. The model is headed to the same place — a utility you draw on, not a differentiator you worship.
What is not converging is context. The model is a general engine that knows everything in general and nothing about you in particular. It becomes useful — becomes yours — only when it is pointed at your material: your email, your calendar, your documents, your finances, your health, the texture of your relationships, the decisions you have already made and do not want to re-litigate. That accumulated context is the scarce, non-interchangeable, irreplaceable thing. It is the one asset in the whole stack that cannot be downloaded, re-trained, or competed into a commodity, because there is exactly one of it and it is yours.
The model is the engine. Your context is the fuel. Everyone is fighting over the engine while quietly planning to own the fuel. — where the value actually sits
So the real question is custody
Once you see that the value has moved from the model to the data layer, the only question that matters is a question about location. Where does the context live? Whoever holds that layer holds the value — and holds you, because an assistant that knows your whole life, running on someone else's servers, is a dependency you cannot easily walk away from. The convenient answer, the one being built at enormous scale, is that your context should pool on a platform's machines, where it can be read, mined, cross-referenced, and monetized in ways you will never see. The model is free; the harvesting is the business.
There is another answer, and it is not a softer version of the same thing. It is to keep the data layer on hardware you own. Bring whatever model you like — rent it, run it, swap it out next quarter when a better commodity ships — but let it reach into your context from the outside, under your consent, without ever taking a copy home. The model visits; it does not move in. Your context stays a place you inhabit rather than an asset someone else holds in escrow against your continued subscription.
Ownership of the layer that matters
“Own your data” has been a slogan for so long that it has gone soft, so it is worth being precise about what it can mean now. It does not mean you get an export button and a cheerful privacy policy. It means the layer that holds the actual value — the context that makes any model useful — is in your custody, on your machine, and the guarantee that no one is reading it is a fact about where it lives rather than a promise about how nicely it will be treated. That is a different kind of ownership than the industry is offering, and it is the only kind that survives the model becoming a utility.
So let the model-builders have their arms race. Let the parameter counts climb and the benchmarks fall. None of it settles the question that will decide how this era actually feels to live inside, which is not how smart is the engine but who owns the fuel. Commoditize the model — please, commoditize it all the way down to the price of electricity. Just do not let the same move quietly hand away the one thing that was never a commodity to begin with. The model is a thing anyone can build. Your life is not.
This is early thinking, and the shape of the data layer is still being poured while we argue about it. But the fork is already here, and it is worth naming before it sets: the model is a commodity, your context is not, and the whole game is being decided right now over who gets to keep it.