The ‘Current’ Situation: The Grid Can’t Take the Weight of Israel’s AI Ambition
Israel wants 100,000 AI processors built on its own soil, plus a national quantum computer. Its own regulator just froze the queue for the power to run any of it.
On July 21, the Electricity Authority and Noga, the state grid operator, stopped processing new data center connection applications of 8 MW or more. The freeze runs 140 days, to roughly early December. The trigger: in two months, new connection requests hit 19,000 MW. Add the existing 8,000 MW backlog and total pending demand reached 27,000 MW, against a country that runs on 9,000 MW average and peaked at 17,000 MW last August, on the hottest day of the year.
One applicant, Mega Or, filed for close to 14,000 MW on its own, roughly half of everything pending. Noga’s own VP of planning, Uzi Zrahia, put it plainly: the system has already committed over 1,500 MW to server farms, and “we used up the capacity of the system until 2035.” A senior industry official went further to Calcalist: “These are insane numbers, and there is no scenario in which we can catch up with such demand.”
The flood didn’t come from nowhere. In February, the government moved to classify large data centers as national infrastructure, the same legal category as ports and airports, letting them skip municipal zoning fights for a fast national approval track instead. The bill defines a qualifying data center as one drawing 50 MW or more, and caps the fast track at 10 new server-farm plans a year nationally, aimed at pushing development into the Negev and away from the already-congested center. That bill has not finished passing. The law meant to make data centers easier to build hasn’t fully passed yet, and the applications already broke the queue.
Underneath the freeze sits a sector that likes to describe itself as strategically inevitable. Startup Nation Central’s own EnergyTech report, out this year, counts 369 active Israeli energy companies, up 68% since 2014, and a record $1.7 billion in early-2026 M&A. Two acquisitions account for the entire number: DustPhotonics to Credo Technology, and LocusView. That’s more than triple all of 2025’s disclosed M&A, from two deals.
Look at who the report names as the sector’s standouts and the pattern is not energy innovation, it’s defense-tech redirected at a new customer. ZutaCore, SaiFlow, Siga, SolidT: cooling systems and grid-cyber protection built by people who came out of Israel’s cyber and defense base, now selling into AI data centers because that’s where the demand moved.
Not everything in the report however is defense-tech in a new label. QDM is working on room-temperature superconductor technology, citing Nobel laureate research, aimed at eliminating transmission losses entirely, real materials physics, and the kind of claim that deserves skepticism precisely because room-temperature superconductivity has burned people before. H2Pro is running actual electrochemical hydrogen production at a 5 MW solar-powered site in Spain with Doral Hydrogen. Phinergy is building aluminum-air batteries for data center backup, a different battery chemistry from the lithium-ion default. Airovation is mineralizing captured CO2 into construction material, solving carbon capture and cement feedstock at once.
But the report itself keeps using the words dual-use and defense-tech to describe the burgeoning energy-tech sector. The regulatory response follows the same logic, and it’s mostly about triage, not generation. The Electricity Authority is drafting rules requiring developers to pay annual holding fees and prove financing before they can hold a place in the queue, because officials found many applicants had reserved capacity with no confirmed land or funding behind it. A joint committee of the Ministry of Energy, the Electricity Authority, and Noga is writing a national allocation policy to rank AI data centers against residential growth, industry, and EV charging. None of this adds a single megawatt. It just decides who waits.
The generation side is thinner than the ambition requires. Natural gas already supplies over 70% of Israel’s electricity, and climbing, as the last coal plants at Orot Rabin and Rutenberg get phased out. An Energy Ministry committee is reviewing whether to keep exporting gas from Leviathan and Karish or hold more of it back for domestic power as data center demand climbs, a decision that pits export revenue against grid security. Solar is targeted to reach 30% of electricity by 2030, which means growing capacity to roughly 17 GW in a country with limited land and no hydro or nuclear to fall back on. Noga is running a 17-billion-shekel ($4.5 billion) transmission and substation build-out through 2030. The Electricity Authority’s own admission, on the record: 80% of grid projects are already behind schedule. That’s an execution shortfall the Authority is confirming about itself, before the AI surge is even fully priced into the plan.
There are some other ideas floating around. A 40% renewable self-supply requirement for large data centers, developers generating part of their own power on-site, was floated by Dr. Gideon Friedman, CTO of NetZero Tech Ventures and a former Energy Ministry chief scientist, as his own recommendation. It is not adopted regulation. What regulators are actually weighing is narrower and still in draft: financial holding fees to screen out speculative applicants, and the ability to curtail farms at peak-demand hours.
Israel has done this before. A country with an isolated grid, no fuel imports to lean on, and a small domestic market has spent three decades turning resource constraint into an engineering discipline. Offshore gas discoveries a decade ago solved an energy security problem the same way; the current freeze could be the same pattern; capacity catches up, the backlog clears, and in three years this reads as a temporary chokepoint on the way to scale rather than a structural limit. The Ministry of Energy’s own chief scientist, Dr. Brian Rosen, frames the sector’s job as diversification across dual-use solar, hydrogen, geothermal, nuclear, and a modernized grid, all at once. That’s a real strategy.
But the strategy assumes time the numbers don’t currently support. The freeze runs to December. The pending request queue is three times national demand. Grid projects are already 80% behind schedule before this surge hit. At least one company has already walked: VisionWave abandoned a planned hyperscale data center joint venture in Israel, naming the freeze as the reason.
The National AI Plan Israel’s cabinet approved on June 16, roughly 2.6 billion shekels of initial investment, targets 100,000 AI processors and a national quantum computer. It treats compute and capital as the constraint. The grid data says the real constraint is megawatts, and right now Israel is allocating them by building a queue, not by adding supply fast enough to clear it.
Sources
Israel’s AI ambitions hit an electricity bottleneck as data center demand explodes — Calcalist/CTech
Regulator makes future of server farms in Israel uncertain — Globes English
AI and the threat to power grid stability — Jerusalem Post
Israeli Electrical Authority freezes grid processing of new data centers for 140 days — DataCenterDynamics
VisionWave Abandons Israeli Hyperscale Data Center Joint Venture — TipRanks
תזכיר חוק: חוות שרתים לבינה מלאכותית – תשתית לאומית — law.co.il
תכנון במסלול מואץ בוות״ל — Nadlan Center
אושר בקריאה ראשונה: חוות שרתים ל-AI תוגדר כתשתית לאומית — C14
חרף החשש מפגיעה בצרכני החשמל — חוות שרתים יזכו למסלול מהיר גם במרכז הארץ — TheMarker
הממשלה אישרה תוכנית לאומית לבינה מלאכותית — Ynet
התוכנית הלאומית החדשה ל-AI: ישראל מכוונת ל-100 אלף יחידות עיבוד — Israel Defense
Israel Energy Country Commercial Guide — US International Trade Administration
2026 Israeli EnergyTech Map Spotlight — Startup Nation Central
Thanks for your continued writing on this, it’s very informative.
Exactly the same problem in the UK, but with less sunshine