On 9 November 2026, Brig. Gen. (res.) Erez Askal, who heads the National AI Directorate in the Prime Minister's Office, told AI & Cyber Week at Tel Aviv University that Israel must build its own AI infrastructure. His reasoning, as quoted by CTech: "If it's not your AI, you have no control over your country." The target is a base of around 300,000 GPUs. About 100,000 would serve domestic users in academia, government, defense and startups, and another 200,000 would serve technology companies and data centers. Askal also said Israel needs roughly two gigawatts of electricity capacity for AI infrastructure, data centers and chip manufacturing.
The strongest case for the plan
The sovereignty argument deserves a fair hearing. Askal warned that within five to seven years, essential systems such as energy, water and healthcare will rely increasingly on AI. A country that runs those systems on models and compute it cannot inspect, secure or guarantee access to has a real strategic exposure. Israel's security situation makes supply-chain leverage a practical concern rather than an abstraction. A state that wants assured access to compute in a crisis has a legitimate reason to own some of it.
The plan is also less autarkic than the headline suggests. Askal said Israel needs its own frontier model but "we do not have to build it alone," and Ynet reports that this could involve allied countries. The strategy, per CTech, does not call for Israel to develop everything independently. That is the right instinct: sovereignty defined as assured access and the ability to audit is achievable, while sovereignty defined as building every layer is not.
What is still missing
The central problem is that the numbers have no financing attached. Ynet reports that Askal gave neither a budget nor a timetable for the targets in his presentation. CTech likewise says the goals "will require substantial investment and infrastructure development" without a figure.
The scale gap is large. In a critical piece on the Directorate's August plan, Globes described a statement of intent to buy up to 100,000 Nvidia processors "within a few years at an undisclosed cost." It said the current national supercomputer has about 4,000 processors, leased from Nebius. Globes also noted the plan had no itemized budget. The Directorate began small: the Jerusalem Post reported on 16 June that Askal was given NIS 120 million, mostly for 2026, and 20 positions. Moving from roughly 4,000 processors to 300,000 is a jump of about two orders of magnitude, and the new figure is three times the roughly 100,000-GPU goal reported at the June cabinet decision.
Power is the harder constraint. Two gigawatts is a very large load for a small country with a constrained grid and limited land for data centers. Globes reports that the Directorate is working with the Ministry of Finance and the Electricity Authority on new criteria for granting electricity approvals, and that industry criticizes the approach as reactive. If grid connection and permitting are the bottleneck, then buying chips first is the wrong sequence. Faster interconnection approvals, transparent queue rules and private generation would do more for real capacity than any procurement announcement.
Regulation: the promise and the gap
The Jerusalem Post reported in June that the Directorate was expected to help Israel pursue regulation "to create order in this new area, while avoiding over regulation which stifles innovation." That is the right goal and it matches Israel's existing approach. Israel has no codified horizontal AI statute. Its December 2023 policy from the Ministry of Innovation, Science and Technology, developed with the Ministry of Justice, favors sector-specific regulators applying principle-based, risk-based oversight, and leaves open the option of horizontal legislation later. Sectoral oversight lets financial, health and other regulators match rules to actual risks rather than imposing a single template. For background on the national programs, the Knesset Research and Information Center has published a comparative summary of national AI strategies, including Israel's.
The weakness is that none of the reporting on the November speech describes any concrete regulatory proposal. A sovereignty-driven agenda carries a specific risk here. A state that owns or subsidizes the compute, runs the national model and sets the rules has strong incentives to favor the champions it built. Light-touch regulation is credible only if it comes with safeguards: open access terms for the 100,000 domestic GPUs so startups and universities are not rationed by agency preference, published criteria for who gets compute, and independent review of any national-model procurement.
The dual role also matters for speech. The Directorate's priorities include defending against deepfakes. Targeted rules on fraud and election deception are defensible, but they should be narrow, and the Directorate should not become a de facto content regulator through its control of national infrastructure.
What to watch
- Funding. Whether the 300,000-GPU figure appears in a budget line or a cabinet resolution with dates, or remains an aspiration.
- Power. Whether the Electricity Authority's approval criteria and any grid build-out actually add capacity on a timeline that matches chip deliveries.
- Access rules. Whether domestic compute is allocated through transparent, contestable criteria.
- Regulation. Whether the Directorate publishes a concrete, risk-based framework, or whether "avoiding over-regulation" stays a slogan.
The diagnosis is sound: dependence on foreign compute is a real vulnerability. But a target without a budget, a timetable or a power plan is a statement of intent, not a strategy. Israel's strongest move would be to publish the financing and the allocation rules before adding more chips to the headline.