ONE SUPPLIER

The real news

The story, straight

Every country says it is building sovereign AI. One company is selling all of them the track.

Sovereign AI means the AI a government runs answers to its own laws, in its own language, on its own data, and keeps working if a foreign company switches it off. If it can be switched off from outside the border, it is a rental, not a utility.

A country needs four layers: power, chips, models and people. Most buy the first two, rent the third and never own the fourth.

Sovereignty is a dial, not a switch.

Dubai is the story everyone repeats. The UAE became the first country to enable ChatGPT nationwide under Stargate UAE, a 1GW Abu Dhabi cluster with 200MW due in 2026.

"Free Plus for every citizen" was reported everywhere and never confirmed by OpenAI: the deal is access through government, schools and health, not a subscription in every pocket.

The schools move the other way. New York City banned student-facing AI through grade eight for a year, ~600,000 students, and banned companion chatbots in every grade.

Los Angeles blocked it district-wide; Katy ISD in Texas banned it from kindergarten to grade six; Ontario's Bill 137 sets up a school ban and Manitoba has said the same.

The fears are real, but the arithmetic is brutal: sovereignty is a generation that can build and judge these systems, and that generation is being raised on a banned tool.

Who is first? Nvidia.

About 92% of sovereign AI model projects run on its chips, and sovereign AI is ~14% of its revenue (~$30B).

The counterweight is open weights: Chinese labs ship models a country can run and retune on its own power, the cheapest real sovereignty on the table. Anthropic filed to go public at up to two trillion dollars without a profit, the expensive version of the opposite bet.

Sources: OpenAI; Gulf News and Benzinga; AP and Education Week; Houston Public Media; Ontario Bill 137; CBC; Stanford HAI; Aju Press; NPR and CNBC TV18; Moonshots 306.