Noga, the company that operates Israel's electricity system, published the list of requests to connect data centres to the grid. Around one hundred requests, totalling roughly 27,000 megawatts.
To grasp the order of magnitude: that is almost three times the country's average consumption, and roughly its entire installed generating capacity.
Errors or inaccuracies are possible. Spotted something that looks wrong? Write to me and I will correct it.
The Numbers, in Proportion
| Total requests | about 27,000 megawatts |
| Of which, arriving within two months | about 19,000 megawatts |
| Israel's average consumption | about 9,000 megawatts |
| Peak demand, August 2025 | about 17,000 megawatts |
| Installed capacity, end-2025 | about 27.2 gigawatts |
That is, had every request been approved, Israel would have had to double its electricity system - and then some.
And nineteen gigawatts of them arrived within two months. That is not a wave. It is an avalanche.
And the Sentence That Answers the Question
The Electricity Authority phrased the answer in terms that admit no second reading:
Existing commitments to connect data centres, amounting to about 1.5 gigawatts, already consume the entire planned capacity addition of the electricity sector through 2035.
Read that again.
Not 27 gigawatts. One and a half. And even those - not out of a surplus, but out of everything planned to be built over the coming decade.
| What was requested | 27,000 megawatts |
| What is already committed, and exhausted the plan to 2035 | 1,500 megawatts |
| The ratio | about 5.6% |
That is the grid's absorption capacity, in a single number.
And so on 26 August the Authority announced a freeze:
| The freeze | |
|---|---|
| Duration | 140 days |
| Applies to | new requests of 8 megawatts and above |
| Until when | early December 2026 |
In the wording of the announcement: "Because of the extreme development in the scope of applications for connection to server farms... the Electricity Authority is announcing a halt to responses and processing of applications for connecting server farms for 140 days."
And during that period the Authority will re-examine the assumptions by which requests are assessed: supply reliability, system redundancy, competition in the electricity market, future electricity prices and natural gas consumption.
Why This Is More Severe in Israel Than Anywhere Else
I wrote a week ago that a grid connection is the global bottleneck of AI infrastructure - four to seven years of waiting, against two to three to build a campus.
In Israel there is an additional factor absent in Europe or the United States: the country is an electrical island.
It has no electrical interconnection to neighbouring grids. A German grid running short imports from France; the Texas grid is tied to its neighbours; Israel is connected to no one.
The practical meaning: every megawatt an Israeli data centre consumes must be generated inside Israel, at that same moment. There is no backup, no import at peak, and nobody to absorb a planning error.
Which is why the grid's reserve margin is not an engineering convenience - it is a condition of functioning. And that is precisely what the Authority says it is protecting.
And on top of that, peak demand is expected to rise 30% to 40% by 2035 even without data centres - from population growth, the electrification of transport and industry, and extreme weather events.
But How Much of the 27 Gigawatts Is Actually Demand
And here one has to stop, because this is the question that decides whether this is a shortage or a queueing bubble.
The Authority itself said that during the freeze it will examine "project credibility". And that is polite phrasing for a familiar problem: when a place in the queue is worth money, it pays to file a request even without a funded project behind it.
The applicant's logic is simple: a request costs little, and a place in the queue is worth a lot.
Anyone holding industrial land beside high-voltage infrastructure can file for 500 megawatts and hold an option - if a partner appears, there is a connection; if not, almost nothing was lost.
So a list of 27 gigawatts is not 27 gigawatts of demand. It is a mixture of funded projects, genuine intentions, and speculative options - and nobody knows the mix, including the Authority.
Which is exactly the reason for the freeze. Not to stop demand, but to separate the types of request before allocating a resource there is not enough of.
And the size of the requests themselves reinforces that reading. The largest on the list, 1,111 megawatts at the former Alliance site in Hadera, is on its own larger than all the commitments already approved across the system. Requests of 500 megawatts recur again and again - in Beersheba, Kiryat Gat, Kiryat Shmona, Tiberias and Lehavim.
A 500-megawatt data centre is among the largest in the world. That several such should exist in a country the size of Israel, simultaneously, is not a plausible scenario.
Who Benefits From This
And here the logic is simple: when a resource becomes scarce, whoever already holds it gains.
The first - whoever already holds an approved connection.
Those 1.5 gigawatts of existing commitments became a different asset this morning. Anyone who obtained a connection permit before the freeze holds something that cannot currently be obtained at any price - at least until December, and in practice well beyond it.
And this is a point worth internalising: in a market like this, the permit is worth more than the land and the building.
The second - the independent power producers.
When demand exceeds supply, whoever generates sets the price. Private producers already account for 61.7% of connected generating capacity in Israel, against 38.3% for the Israel Electric Corporation.
I covered several of them in the recent reporting season - Dalia Energy, Nofar Energy, Doral, Prime Energy and Generation Capital - and what they had in common was high leverage and sensitivity to financing. A scarcity environment favours their revenue side.
The third - owners of industrial land beside infrastructure.
The list contains the names of income-producing property and retail companies: Mega Or, Big at several sites, Iskur, Kargal, Sano. Their land did not change designation and did not grow - it simply sits beside a transmission line.
And that is the kind of thing that changes value without anything being done to it.
And the fourth - gas suppliers and infrastructure.
The Authority itself warns that approving further commitments would "increase natural gas consumption beyond previous forecasts". That is a regulator's statement about future gas demand.
And alongside it - whoever builds the grid. Substations, high-voltage lines and electrical infrastructure are the only way to open the bottleneck, and they are infrastructure contractors' work.
And Who Is Hurt
The first - most of the list.
If the grid can absorb 1.5 gigawatts and 27 were requested, then about 94% of the requests will not be connected in any foreseeable timeframe. Some will fall because they were speculative to begin with. Some will fall despite being real.
The second - whoever was late.
The freeze blocks new requests above 8 megawatts until December. And anyone filing afterwards will stand in the queue behind everyone already inside - including behind the speculative requests not yet filtered out.
The third, and this is the least discussed - the industrial consumer.
A data centre does not only compete with another data centre. It competes with a factory. The same megawatt, the same grid.
And the Authority says so explicitly: approving further commitments could "reduce the reserves needed for natural economic growth" and ultimately "affect electricity prices for consumers".
That is, the regulator itself signals that the price of this wave, if answered in full, will be paid by those who asked for nothing.
And the fourth - Israel's sovereign AI ambitions.
In the piece on Nvidia we saw that "sovereign customers" are one of its fastest-growing engines - states building national compute infrastructure. A country that cannot connect a megawatt cannot be such a customer.
What I Will Follow
| Early December | The end of the freeze - and the new allocation rules |
| The credibility test | How much of the 27 gigawatts survives a screen for funded projects |
| Industrial electricity prices | The Authority flagged this explicitly |
| Generation plans | Whether the plan to 2035 is revised upward |
| On-site generation | Private gas turbines as a way around the grid |
| Already-approved connections | Which of them actually progress to construction |
הזווית שלי
דעה אישית של אילן אברמוב - לא ייעוץ ולא המלצה
What interests me here is that Israel has received, in concentrated and extreme form, the same problem I wrote about in the global context - and it is easier to see precisely because of the scale.
In a small country that is also an electrical island, there is nowhere to hide. In the United States you can build in Ohio instead of Virginia. In Israel there is no "somewhere else" - there is one grid, and every megawatt comes from it.
And the number I take from this day is the ratio: 1.5 against 27. Not because it is precise - it will change - but because it defines the type of problem. This is not a problem of prioritising between better and worse projects. It is a problem in which almost everything requested cannot happen.
And what I think matters most for the reader is the distinction between the two: between demand and a queue. A list of 27 gigawatts looks like enormous demand, and it is mostly options - places in a queue bought cheaply. Anyone reading the number as though it describes funded projects will misjudge the size of the market several times over.
And that is also why I think the freeze is right, and not because it stops AI. It stops the queue. When a scarce resource is allocated by order of arrival rather than by feasibility, it goes to whoever was fastest, not to whoever will build.
And here I want to be fair to the other side of the argument. There is a reasonable reading that says a 140-day freeze is exactly what drives investment away: anyone planning a data centre costing tens of millions of dollars does not wait for a regulator, they go to Cyprus or Greece. And when an industry is dividing itself between countries precisely now, four months is a long time. That argument is strong and I do not dismiss it.
And what I will really be looking for in December is not whether the freeze was lifted, but by what rule connections are allocated afterwards. If the rule is "first come" - we are back where we started. If it is conditional on proof of financing, on construction progress or on on-site generation - that is a rule that separates a queue from demand.
And the difference between those two rules is worth more than any megawatt on the list.
(It is important to stress: this is my personal opinion only, and nothing herein constitutes a recommendation to take any action.)
