Jensen Huang: every nation must own its AI production

Sovereign AI entered the vocabulary through NVIDIA's Jensen Huang, who told the February 2024 World Government Summit that every country needs to manage the production of its own intelligence. The framing has since been filled out into four layers — data, models, computing infrastructure, and operations and people — which together describe who decides how AI is used, who manages it, and who responds when something goes wrong.

The distinction that matters is between where data sits and who controls it. A domestic data center managed by a foreign company is not sovereign in this sense, and the U.S. CLOUD Act is cited as an example of how legal jurisdiction can reach data stored abroad. Likewise, a domestic LLM can still rely on overseas cloud services, while an open-weight model from abroad can raise autonomy if a country operates and improves it locally.

Countries are already testing the concept. Denmark's Gefion supercomputer supports weather-forecasting AI, Switzerland's Apertus discloses weights, training data, and development steps, and Japan is building domestic infrastructure alongside its 2025 AI Act and AI Basic Plan, with the Digital Agency planning 2026 trials of domestic LLMs on its Gen'nai environment. The practical question, as the piece frames it, is not full self-sufficiency but which parts a country or organization must keep under its own control.