From Watson to Sovereign AI

The Question

Every artificial intelligence system ever built was born from the same question: how does the machine know what to do next? I asked that question at age three, staring at a Commodore VIC-20 my brother had brought home, and I have spent thirty-five years following the answer. When I first saw Sid Meier's Civilization — a machine making strategic decisions across an entire simulated world — I knew the answer was not simple. The machine was not following a script. It was evaluating, adapting, choosing.

Words and Logic Become Cheap

I was studying Canadian law and political science when IBM's Watson defeated the best human Jeopardy players. The implications were immediate: if a machine can process natural language well enough to compete at Jeopardy, then every profession built on words and logic is living on borrowed time. Legal research. Policy analysis. Financial modelling. The entire architecture of white-collar employment was sitting on a trapdoor. I left university and spent a decade learning twenty percent of every major trade — electrical, networking, server administration, carpentry. With twenty percent of each trade, you can do eighty percent of the work. I bet that when the machines came for the knowledge workers, the builders would still be standing. That bet is paying off.

Tokenization: Boring, Dramatic, Transformative

The decade between Watson and ChatGPT was a slow burn for anyone paying attention. Tokenization — the process of decomposing language into discrete computable units — advanced from academic curiosity to the foundation of the most powerful AI systems on Earth. It is often called the most boring end of artificial intelligence. It has nonetheless produced the most dramatic results. The computational and electrical costs are staggering, but the results speak for themselves and will only keep improving. When you apply enough computation to enough tokens, something unexpected emerges. The machine does not just match patterns — it reasons, infers, creates.

Sovereignty Is an Infrastructure Problem

The insight that drove me to found Yamoria is this: whoever controls the compute controls the AI. Training frontier models requires thousands of GPUs running for months. Inference demands sustained capacity indefinitely. This is not a software problem — it is an infrastructure problem. And right now, that infrastructure is overwhelmingly controlled by American companies under American law. Canada has a choice: build sovereign compute on Canadian soil, or accept permanent dependency for the most consequential technology of this century. I chose to build.