Israel is drawing up plans to build its own large-scale artificial intelligence infrastructure, including approximately 300,000 GPUs hosted inside the country and the development of an advanced Israeli foundation model, according to remarks by the head of the government's National AI Directorate reported by Israel Defense and CTech.

Brig. Gen. (Res.) Erez Eskel, who leads the National AI Directorate in the Prime Minister's Office, presented the strategy at the Cyber Week conference and an AI conference held at Tel Aviv University, arguing that national independence in AI is becoming inseparable from national independence itself.

Why Israel Wants Its Own AI Stack

Eskel warned that within five to seven years, essential systems across the energy, water and healthcare sectors are expected to rely increasingly on artificial intelligence. If those systems run on models and infrastructure that Israel does not control, he argued, the country's sovereignty is at stake.

"If a country's critical infrastructure operates on the basis of artificial intelligence that it does not control, it will not have full control over its own country," Eskel said, according to Israel Defense.

He also pointed to the limits of relying on foreign vendors: commercial companies act according to their own business interests, which do not necessarily align with those of the state, and cannot guarantee continued access to their services and models. In his view, compute should be treated the same way a country treats its electricity grid or water supply — as critical national infrastructure that cannot simply be outsourced.

The timetable matters as much as the target. A five-to-seven-year window in which hospitals, utilities and transport networks grow dependent on AI leaves governments little room to renegotiate their position once the dependency is in place — which is why the directorate is pressing for capacity to be built now, while the technology's role in critical services is still being decided.

The Numbers: 300,000 GPUs and 2 Gigawatts

The plan's computing target is split into two tiers:

  • Approximately 100,000 GPUs reserved for government, the defense establishment, academia and startups
  • Approximately 200,000 GPUs to support the growth of Israeli technology companies and data centers

Delivering that capacity on domestic soil will require an estimated two gigawatts of electricity — a significant figure for a country of Israel's size, and one that puts the plan in the same conversation as the power-hungry data center buildouts reshaping national grids in the United States and Europe.

CTech, which first reported the scale of the GPU target, framed the initiative as Israel's bid to build its own AI infrastructure rather than renting intelligence capacity from abroad.

An Israeli Foundation Model

The second pillar of the strategy is the development of an advanced Israeli foundation model — a native AI system that would allow the state to operate AI-powered services without becoming entirely dependent on cloud platforms and models developed by foreign companies.

Eskel indicated the effort could also be advanced in cooperation with countries facing similar sovereignty challenges, signaling a possible bloc of mid-sized nations building alternatives to the handful of US and Chinese labs that currently dominate frontier model development.

Beyond Sovereignty: Public Services and Defense AI

Alongside the technological independence agenda, the strategy outlines plans to apply AI to improving public services in healthcare, education and transportation, while reinforcing Israel's existing strongholds in cybersecurity, defense and the emerging field of physical AI.

"We need to talk about artificial intelligence not only in technological terms, but also ask what kind of world we want to live in," Eskel concluded. "Countries that take action will maintain control over their future. Israel has decided to act."

Part of a Global Sovereignty Race

Israel's plan lands in the middle of a worldwide scramble for AI self-sufficiency. The European Union has tied its AI strategy to home-grown gigafactory compute projects, Gulf states have committed tens of billions of dollars to domestic AI infrastructure, and Japan, South Korea and India have each announced sovereign compute programs of their own.

The pattern is consistent: nations that watched a small group of US companies control the frontier models powering the current AI boom are now willing to spend heavily — on chips, on power and on indigenous models — to avoid permanent dependency.

For a country whose economy leans heavily on high-tech exports and whose defense sector is intertwined with commercial technology, the calculus is particularly acute. The open questions are execution ones: securing hundreds of thousands of GPUs in a supply chain where demand far outstrips supply, wiring two gigawatts of generation capacity into the plan, and funding a foundation model capable of competing with systems trained at far greater scale.

None of those constraints, however, change the direction of travel. As Eskel framed it, the choice for governments is no longer whether critical systems will run on AI — it is whether the AI they run on will be theirs.

For more context on this story, see our ongoing AI policy coverage.

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →