'Sovereign AI' is an umbrella term for a country's ability to develop, run and control AI without depending entirely on foreign companies or governments. It does not mean a country must build everything itself, but that it needs three components under its control: compute, meaning data centers and processors; models, meaning AI systems trained and run under its rules; and data, kept within its borders.

Why does it matter? Imagine a government ministry, a hospital or the military using an AI model from an American company running on servers abroad. If the company changes its terms, raises prices, or the US government restricts use, the country is left without a tool that has become part of its infrastructure. In addition, models trained mostly on English and American culture do not always understand small languages like Hebrew, local laws or cultural context well.

The expensive part is compute. Training a large model requires tens of thousands of graphics processors, each costing tens of thousands of dollars, and data centers that consume huge amounts of power and water for cooling. That is why countries talking about sovereign AI talk first about processors and gigawatts, and only then about models.

Who is already doing it? France, Britain, Japan, India, the UAE and Saudi Arabia have announced multibillion-dollar national programs, some in partnership with Nvidia, which has promoted the 'sovereign AI' concept worldwide in recent years, not without self-interest: every country building a national data center is a new customer. In Israel, the head of the National AI Directorate, Erez Askal, presented targets this week of 2 gigawatts and hundreds of thousands of processors.

Is it worth the money? The answer depends on what the country does with the infrastructure. A national data center serving researchers, startups and public systems at a fair price can become a competitive advantage for years. One built for PR that stands half empty, or becomes obsolete before construction ends, is an especially expensive way to waste public money. In a field where a new generation of chips comes out every year, timing is also part of sovereignty.