A government agency we work with was once offered an AI trial at almost no cost — on one condition: the documents they’d feed in could be processed in a data centre on another continent. The trial would have worked. It would also have created a permanent dependence, a permanent legal exposure, and a precedent that later contracts couldn’t easily undo. They said no. They were right, and this is an attempt to explain why with more than instinct.

The instinct is easy to dismiss as flag-waving. It isn’t. The case for keeping things at home stands on three plain grounds — control, legal exposure, and the long-run cost of depending on someone else — and it holds up on each one separately.

What “at home” means

“In your country” is often reduced to where a database sits. It’s more than that. There’s where the data is stored, where it’s actually processed while the AI is working, and who holds the contracts that govern all of it. A setup can be half-sovereign — stored at home, processed abroad — and that half-answer is often the one quietly delivered. We treat the whole thing as the unit: stored and processed inside the country, on systems you control, under contracts with a local company, with the AI providers barred from training on your documents.

Sovereignty isn’t where your data sleeps. It’s where your data is when it’s being used — and who can demand to see it.

The trap in “free”

The cheap cross-border trial is tempting because it removes the hardest part — setting up local infrastructure — at no apparent cost. The cost is just deferred, and paid in dependence. Once your documents, your habits and your staff are built around something you don’t control, the cost of moving rises every year. The trial that was free to start becomes expensive to leave.

The legal side

For regulated organisations this isn’t only strategy; it’s law. Nigeria’s data-protection rules (NDPR and the NDPA 2023) put real limits on how personal and institutional data is handled and where it can go. A cross-border setup can pass a slick demo and quietly fail a data-protection review. We’d rather have the awkward conversation before the setup is chosen than after something goes wrong — so our default is at home, no training on your data, with the privacy review done early.

It’s also just better

The argument is usually framed around politics and compliance. It’s worth ending on the part that’s just engineering. Processing at home is faster for local users. A setup you control is easier to secure and easier to make sense of when something breaks. And the unchangeable record we keep is far easier to stand behind when the whole thing sits in one country, under one set of contracts.

None of this is an argument against the big AI models built elsewhere — we use them, under contracts that keep your documents in place and untouched. It’s an argument about where the system around them lives, and who’s in charge of it. For African organisations, the answer that holds up on control, on law, and on plain engineering is the same: at home.

Back to the index

More Notes as we have something useful to say.

The full list, plus the pieces still in the queue.

All Notes →
Talk to the assistant

“Your data stays yours” is one of four promises.

The design agent on the home page sketches a private AI workbench for your business and emails the plan to you.

Talk to the assistant →