Israel sets 300,000-GPU target to reduce reliance on foreign AI providers

Israel wants the AI behind its essential services to be less dependent on foreign technology companies. On November 9, National AI Directorate chief Erez Askal presented a strategy targeting about 300,000 GPUs, roughly two gigawatts of electricity capacity and a domestic foundation model, according to CTech. These are planned capabilities, not completed deployments.

Speaking at AI & Cyber Week at Tel Aviv University, Askal tied the decision to an expected change in critical infrastructure. Within five to seven years, he said, energy, water and healthcare systems would increasingly rely on AI. His concern is who controls the technology those services will need.

A computing target with an electricity requirement

The proposed hardware base would serve both domestic institutions and commercial infrastructure. The government’s target divides the roughly 300,000 GPUs into two groups, rather than assigning the entire fleet to government work or to developing a single national model:

• About 100,000 GPUs would serve domestic needs across academia, government, defense and startups.
• Another 200,000 GPUs would support technology companies and data centers.

Askal put the electricity requirement at approximately two gigawatts. That figure covers AI infrastructure, data centers and semiconductor manufacturing; it is not a power allocation solely for the proposed GPU base. The strategy therefore reaches beyond acquiring computing hardware to the electricity capacity needed to support the planned technology infrastructure.

Delivering those targets will require substantial investment and infrastructure development. CTech noted that Israel’s electricity grid is already under growing pressure from data-center demand. That constraint leaves the speed of the proposed expansion an open question, even as the strategy establishes the scale the government wants to reach.

Domestic control does not mean building alone

The model effort addresses a different layer of dependence. Askal argued that leading AI models are largely available through external cloud services. That makes it difficult, he said, to run them within a country’s own systems while fully protecting sensitive information. An advanced Israeli foundation model is a central objective of the strategy.

He also challenged the idea that open-source alternatives settle the issue. In his assessment, those models do not yet match leading systems’ capabilities, and their continued openness cannot be assumed. Those are Askal’s judgments about the available alternatives, underpinning his case for developing a domestic model rather than relying on access to outside systems.

Still, the strategy does not call for Israel to develop every component independently. Askal described an approach that keeps critical parts under national control while working with international partners. Cooperation with countries facing similar challenges, he said, could make the proposed frontier-model project possible.

Beyond computing and models, the strategy seeks improvements in healthcare, education and transportation. It also calls for workforce training and AI skills among younger generations. Askal linked that agenda to questions about employment, education and public trust, presenting the infrastructure effort as part of a wider response to AI’s expected effects on public life.