On June 16, 2026, Israel's government approved Prime Minister Benjamin Netanyahu's National AI Program, a resolution built around a single headline number: 100,000 processing units of sovereign computing capacity, to be made available for government, academic, and industrial AI use. The resolution, issued through the National AI Directorate at the Prime Minister's Office, also commits to a national quantum computer built "to the extent possible" on Israeli technology, workforce retraining across the education system, and a new mechanism to manage AI-driven labor-market disruption.
What the resolution does not do is name a number to pay for its own centerpiece. Tel Aviv University AI researcher Nadav Cohen told Ynet that a 100,000-GPU buildout would cost "$20 billion to $30 billion and potentially more," citing global chip scarcity and the reality that advanced server clusters — the kind built around Nvidia's top-end chips — become obsolete within two to four years, meaning the spending isn't a one-time bill but a recurring one. That figure is an outside analyst's estimate, not a Finance Ministry appropriation. No comparable number appears in the government's own text.
The Case for Centralizing AI Policy
The decision to run this through the Prime Minister's Office rather than the Ministry of Innovation, Science and Technology, which has nominally owned Israeli AI policy since 2021, deserves a fair hearing. Israel created the new National AI Directorate on September 25, 2025, explicitly because the prior arrangement wasn't working. The security establishment's Nagel Committee, cited by Calcalist's CTech in its reporting on the directorate's creation, warned that "Israel today is not in a good position to accelerate the field" while "major countries are investing hundreds of billions." A fragmented, ministry-by-ministry approach to compute, talent, and procurement is a real coordination failure in a field where speed and scale compound. Concentrating authority at the center of government, with direct prime-ministerial backing, is a defensible answer to that failure — it is how Israel has historically moved fast on cross-cutting technology and security priorities.
The State Comptroller's own November 2024 special report backs this diagnosis. It found flatly that "there is no long-term strategy for artificial intelligence in Israel," that no single body coordinated the national program across ministries, and that Israel's standing had slipped on every major index it tracked — the Tortoise AI Index from 5th to 9th globally between 2019 and 2024, the Oxford AI Readiness Index from 20th to 30th. A March 30, 2026 EUROSAI parallel audit coordinated by Israel's own State Comptroller, covering twelve national audit institutions, extended that scrutiny to a comparative European standard. The government is not inventing a problem to justify a power grab; it is responding to its own auditor.
But Ambition Without Appropriation Is a Pattern, Not an Accident
The trouble is that Israel has run this play before. The country's existing five-year national AI program, launched in 2021 with a nominal budget of roughly NIS 5.26 billion, had released only about NIS 1 billion — one-fifth of the total — by April 2025, according to reporting on the program's stalled implementation. No operational national GPU cluster exists yet; a pilot high-performance-computing lab opened in 2024 but remains limited in scale; promised regulatory sandboxes and a foundational AI challenge fund were, in that reporting, "unnamed, unfunded, and absent" from progress updates years after being announced.
Against that backdrop, the resourcing behind the new Directorate is telling. Calcalist's CTech reported the AI Directorate received NIS 120 million (roughly $33 million) for 2026, funded through cross-cutting budget cuts rather than new appropriation, with NIS 13 million annually from 2027. Set beside a $20-30 billion infrastructure ask, that is a rounding error — the announced ambition outruns the appropriated budget by a factor in the hundreds, not the low double digits. Israel is not alone in this gap between AI rhetoric and AI capital: the comparison analysts keep reaching for is the United States' $52 billion CHIPS Act commitment to domestic semiconductor capacity, itself dwarfed by the scale of what frontier AI compute now costs globally, with the EU, UAE, and Saudi Arabia each running their own multibillion-dollar sovereign-compute pushes.
The Proportionate Path
None of this argues against sovereign compute as a goal. A country with Israel's talent density and defense-tech base has a legitimate interest in not being wholly dependent on foreign cloud capacity for AI workloads that touch national security, and centralizing coordination at the PM's Office is a sound structural fix for a genuine failure the Comptroller documented. The problem is sequencing: announcing a 100,000-unit target through a cabinet resolution, without a matching Knesset appropriation or a phased procurement plan that accounts for hardware depreciating every two to four years, invites exactly the outcome the 2021 program produced — headline numbers that don't survive contact with the annual budget process.
The fix is not less ambition; it's ambition sequenced to funding that actually exists. A published multi-year procurement schedule, tied to specific Knesset budget lines rather than departmental reshuffling, would let Israel's tech sector and its allies take the 100,000-unit target seriously instead of filing it alongside the last program's unmet promises.