Local intelligence at the edge
The most interesting AI over the next five years will not be the biggest model in the cloud. It will be the smallest useful model on the machine in front of you.
There is a version of this story where the next decade of AI is a race between three hyperscalers with ever larger models. That version is not wrong, but it is not the whole picture.
The most interesting AI applications we see - in health, in care, in local government, in defence-adjacent industry - cannot use a shared cloud tenant. The data is too sensitive, the regulatory footprint too specific, the connectivity too unreliable, or all three.
That is why we run a Local Intelligence programme. Small, capable models on hardware you can physically point at. Retrieval against corpora that never leave the building. Provenance on every claim. It is less glamorous than the model-of-the-week cycle, and it is where most of the durable value will be built.
Medibase and Symplicare AI are early points on this curve. There will be many more.