Powering Israel’s AI ambitions: the system behind the nuclear debate

An investor’s case for nuclear-powered computing raises a broader question: how should an electricity system serve growing digital demand?

Select an underlined term for a short explanation.

What would it take to make Israel a major centre for artificial intelligence? In an Energya interview covered by Tal Shahaf for Ynetnews, investor Ezra Gardner argues for nuclear-powered computing. The interview describes a proposed Negev technology complex, Fort Foundry, associated with Project Spire. It also identifies his investment connection to reactor developer Terrestrial Energy. Readers are hearing a participant’s case, rather than an independent assessment of the proposal.

Gardner dismisses solar as inadequate for data-centre reliability. The comparison needs a wider frame. The International Energy Agency’s 2025 Energy and AI report examines electricity supplied through entire grids as well as on-site generation. Its base case projects renewables meeting nearly half of the growth in global data-centre electricity demand between 2024 and 2030. That is a modelled contribution to a mixed system, not proof that an isolated solar array can keep servers running continuously.

The agency’s executive summary puts storage and the wider grid alongside renewables, while also assigning a role to controllable generation. It identifies location and flexibility as practical decisions: building where electricity and network capacity are available can reduce pressure. This makes the question more specific than choosing one technology. Which supply, connections and operating arrangements would serve a particular computing campus?

Small modular reactors aim to make nuclear construction more repeatable through factory-made components. The International Atomic Energy Agency describes units of up to 300 megawatts of electrical capacity; fission provides heat used to produce energy. Its explainer cautions that economic competitiveness still needs to be demonstrated in deployment. A technology category therefore does not establish the price or delivery date of a particular installation.

An American example illustrates the stages involved. In March 2026, the US Department of Energy reported construction-permit approval for TerraPower’s Natrium project in Wyoming. Preparatory non-nuclear work had begun in 2024, but operating the reactor requires a separate licence. The department described 2030 completion as an expectation. Permission to build and permission to operate answer different questions.

For the Israeli proposal, SPFI has not verified a signed agreement, an approved reactor design or an operating date. Those gaps remain visible. The useful next evidence would be a documented project and an electricity-supply assessment that readers can examine. Gardner’s ambition opens a discussion; it does not settle what mix of technologies would deliver it.