Infrastructure on loan

The cost argument, the language argument, and the privacy argument are all practical. Each could, in principle, be answered another way: negotiate a better rate, fine-tune harder, redact more. The last argument cannot be answered another way, because it is not practical. It is structural.

When a health system's intelligence runs on a foreign vendor, the vendor holds a lever on the system, whether or not it ever pulls it. The lever is the terms of service: a document the country did not write, that can be changed without its consent, and that governs the layer deciding what care millions of people receive. A single page on a company's website becomes a single point of failure for a nation's health.

Countries already know this instinct elsewhere. They do not run the power grid on a generator they cannot service, or the payment system on rails they cannot govern. Health was slower to see as infrastructure because for so long it was paper and people, with no central thing to depend on or lose. The moment intelligence becomes the thing the system leans on, the same rule applies: the part you lean on is the part you must be able to keep.

This is why owning the model is not a luxury to reach for once everything else works. It is the point at which a clinical AI stops being a clever product and becomes infrastructure a country can build on. A rented mind can be a fine way to start, to prove what works and learn what the country needs. It is not a thing to found a health system on, because the one property infrastructure must have, that it stays, is the one property a rental can never promise.