Rent or Own

There is a question you can ask about any capability that matters: do you own it, or do you rent it? For most of the intelligence we now run on, the honest answer is that we rent it — from about four companies, on terms we don't set and can't see.

It felt like a convenience when it started. You didn't have to run the model; you called an endpoint. You didn't have to hold the data; it lived somewhere warm and managed. But rent has a way of becoming a leash. The price moves. The model changes underneath you. The thing you built your workflow — or your agency, or your court, or your hospital — on top of can be deprecated, rate-limited, or quietly re-aligned, and you find out when it happens to you.

Almost everything I build is an argument with that arrangement. Offline agents that never phone home. Models that run in a room with no route to the internet. A homelab that exists mostly so the things I depend on depend on me back. It looks like paranoia from the outside and it is really just a preference for owning the tools I stake anything on. If the internet went out tomorrow, most people's "AI" goes with it. Mine doesn't. That's the whole point.

At the scale of a person, this is a preference. At the scale of an institution it's a risk decision. At the scale of a country it's sovereignty, in the old and literal sense: the capacity to act without asking permission. A nation that runs its analysis, its language, its decision-support on infrastructure it neither owns nor can inspect has quietly outsourced a piece of its own statecraft to a vendor in another jurisdiction. It works fine right up until the day interests diverge — and then it doesn't, and there is no local fallback, because nobody built one.

I'll give the counter-argument its due, because it's real: the frontier models are better, they're cheaper than doing it yourself, and you'll always be a step behind if you insist on running your own. All true. But "better and cheaper, as long as nothing ever goes wrong with the party you're renting from" is exactly the sentence that precedes every dependency disaster ever written up afterward. Sovereignty has always cost a premium. You pay a little more to hold the thing yourself, and what you buy is the ability to keep operating on the worst day — the day the rented thing is turned off, or turned against you.

I'm not arguing everyone should run a datacentre in their spare room. I'm arguing that can I still do this if the vendor disappears? is a question worth asking before you build your life, your business, or your government on top of an answer — and that the honest answer, for most of the intelligence we now rely on, is no. Owning your intelligence is slower, dumber, and more annoying than renting it. It is also the only version you actually control. In a decade where the thing being centralised is thought itself, that trade is going to look less like a hobby and more like the only sane hedge there was.