An Institute Designed by Its Future Customers
On July 5, 2026, Israel's National AI Directorate sat down at the Manufacturers Association's Industry House with representatives from Oracle, HP, Check Point, Armis, Kaltura, Kyndryl, ELTA, Bynet, CodeValue and HPE — eleven firms in total — for the first in a planned series of consultations ahead of establishing the National AI Institute. The meeting, co-organized with the Israel High-Tech Association, was billed as mapping industry needs and identifying strategic partners before the Institute's design is locked in. It is a sensible instinct: build infrastructure policy with the people who will actually run workloads on it, rather than drafting in a ministerial vacuum. But the venue matters as much as the guest list. Israel is standing up a state AI institution primarily through a government resolution and an industry roundtable — not through legislation that would give the Knesset, competitors left off the invite list, or civil-society reviewers a formal seat.
A Program Built by Resolution
The consultation follows the government's June 16, 2026 approval of Prime Minister Netanyahu's National Artificial Intelligence Program, which sets a target of 100,000 sovereign processing units and, according to industry estimates, could cost $20–30 billion to build out, including roughly 5,000 of the most advanced GPUs earmarked annually for research and industry access from 2027 through 2032, per reporting on the plan's compute allocation (Let's Data Science). The program's pillars — sovereign compute, a national quantum computer, workforce transition, and the National AI Institute itself — sit under the National AI Directorate, a body the Prime Minister's Office created on October 12, 2025 and now runs directly, having pulled AI policy out of the Ministry of Innovation, Science and Technology, reversing a 2021 allocation of that portfolio (Calcalist). The Directorate's director, Brig.-Gen. (res.) Erez Askal, was appointed without a competitive tender; the Civil Service Commission granted an exemption citing urgency and security considerations, explicitly limited to this first appointment. The Directorate's initial 2026 budget is NIS 120 million with 20 new staff positions, settling to roughly NIS 13 million in annual operating costs from 2027 — money that answers to the Prime Minister's Office, not the Innovation Ministry.
The Case for Moving This Fast
There is a real argument for all of this, and it deserves to be stated plainly before it's contested. Israel's most successful piece of tech-industrial policy — the National Cyber Directorate, built under Isaac Ben-Israel's cybersecurity initiative model in the 2010s — also started as an executive-branch creation, iterated quickly with industry input, and only formalized into statute after the institutional model had already proven itself operationally. AI capability, unlike most regulatory subject matter, has a shelf life measured in product cycles: a National AI Institute designed by committee over an 18-month legislative process risks specifying infrastructure for a compute paradigm that's obsolete by the time the law passes. Consulting the ten or eleven firms actually capable of deploying sovereign-scale compute in Israel — rather than starting from a blank regulatory slate — is a legitimately faster way to get infrastructure into the ground.
Where the Shortcut Gets Expensive
The trouble is that the same speed which makes the executive-resolution model attractive also removes the checks that would normally catch its failure modes. The Directorate's own mandate, as described by the PMO, is not narrowly technical — it covers "regulation, infrastructure, and research," alongside strategy formulation (Jerusalem Post). That overlaps directly with the National Cyber Directorate's existing brief on AI-enabled cyber threats, and Israeli cyber officials have already flagged the ambiguity: who leads response when an AI system is the vector for an attack on hospital infrastructure, the INCD or the new Directorate? The INCD's multi-year jurisdictional dispute with the Shin Bet is the closest precedent, and it took years and leadership cross-appointments to settle. Building a second overlapping authority without legislating the boundary first is choosing to relive that fight.
The procurement-adjacent structure compounds the risk. An Institute that will gatekeep access to a scarce, publicly-funded compute pool of 100,000 processing units is, functionally, an allocator of enormous commercial value. Designing that Institute in direct consultation with the same ten or eleven vendors most likely to sell into it, supply it, or operate it is not corrupt — but it is a textbook conflict-of-interest setup that normally gets addressed with published selection criteria, competitive bidding rules, and an appeals mechanism for firms left outside the room. None of that yet exists on the record. A resolution can authorize a budget line; it cannot, by itself, create the kind of durable, appealable allocation rules that make an infrastructure institute trustworthy to the firms not invited to Industry House on July 5.
The Fix Is Legislative, Not Slower
None of this argues for slowing the compute buildout — Israel's AI ambitions are a legitimate response to genuine strategic competition, and the industry-first consultation model beats regulation-first approaches like the EU's that specify compliance obligations before anyone has built the thing being regulated. The fix is narrower: the Knesset should legislate the National AI Institute's mandate, its boundary with the INCD, and transparent compute-allocation criteria in parallel with the operational buildout — not after it. Executive speed and statutory accountability are not substitutes for each other; Israel's cyber-directorate model succeeded specifically because it eventually got both. Skipping the second half this time would trade a genuinely good instinct — build with industry, not around it — for a National AI Institute nobody outside Industry House can hold accountable.