Signal. The IEA put a dedicated grids chapter at the front of its annual flagship for the first time this year. Electricity 2026, published 6 February and released CC BY 4.0, extends the forecast to 2030 and reports one number to read slowly: over 2,500 GW of renewable, large-load and storage projects are currently stalled in grid queues worldwide. Read the noun list out loud. That is not 2,500 GW of stalled wind farms. It is the wind farm, the electrolyser, the steel plant and the data centre all in one queue — “large load” is the hyperscaler’s substation, waiting behind the solar park.

Here is the table the chapter publishes, and the whole piece arrives at it early:

• EV charging — 1–2 years
• Data centres — 1–3 years
• Renewable projects — 1–5 years
• The grid infrastructure they all connect to — 5–15 years

Everything on the fast clock plugs into the thing on the slow clock. An office can design, consent and build a logistics shed with a charging yard in the time it takes the connection application behind it to clear study. That is the news — not a procurement failure, a structural mismatch between two schedules that were never synchronised. The fast clock is the one we are trained to manage; it is not the one that decides whether the building can open.

←TODAY: the queue is live in 2026, and the connection date for anything opening this decade is already fixed — you are only discovering it. →3012: the networks that still run in the Zurich-3012 horizon are the ones whose owners ordered the iron before the permit, and kept a second path. Fulcrum: the connection date only looks like a delay from the fast clock; from the slow clock it was always the real date.

System. The slow clock has a slower clock inside it, and no permit reform can touch it: the hardware. A separate IEA study — Building the Future Transmission Grid, published 25 February 2025 under Alana Rawlins Bilbao with chief energy economist Tim Gould, off a 2024 industry survey — labels the vintage plainly. On those figures it takes two to three years to procure cables and up to four for a large power transformer; direct-current cable, the kind you want for long runs, runs beyond five years. Lead times have almost doubled since 2021, and transformer prices have risen by around 75%.

The order books are the tell. Hitachi Energy’s backlog more than tripled in four years to over USD 30 billion; HD Hyundai Electric and Hyosung Heavy Industries together passed USD 10 billion. And the sentence that carries the whole change, in the industry’s own words: buyers are now “booking a manufacturing slot rather than the traditional model of buying material for delivery.” You are not buying a transformer. You are buying a place in a factory queue sold out to 2029.

The tempting story is that the permit used to be the bottleneck and now it is the hardware. The IEA says the opposite, in the same report: “the primary cause of delays in transmission projects remains permitting, particularly in advanced economies.” An extra-high-voltage line takes on average eight years to develop in Europe and the US, against three in India and one-and-a-half in China. So two independent clocks run at once — an eight-year consenting clock and a four-year iron clock — and they do not run in series. Accelerate the permit and you arrive at the factory queue; expand the factories and you arrive at the objection period.

Which is why the IEA leads its recommendations with visibility of demand and anticipatory investment — someone has to order the iron before the permit is granted. That is exactly the move every procurement rulebook is written to forbid: you may not commit capital to equipment for a project not yet consented. The rulebook written to protect the public purse is the same one that guarantees the queue.

Street. Switzerland reads the same way at home. Swissgrid’s Strategic Grid 2040 — a standing plan set out on 30 April 2025, not fresh news — identifies 31 major projects worth around CHF 5.5 billion by 2040: 400 km of line enhanced, 790 km newly built, and roughly 1,300 km already over 60 years old. The detail that lands it: total transmission grid length stays constant at 6,700 km. Switzerland is not building a bigger grid; it is rebuilding the one it has, under load, while demand rises — and Swissgrid itself names large data centres, alongside heat pumps and electromobility, as the drivers. Set the Swiss operator’s own line beside the IEA’s 5–15 years: “Grid projects often take 15 years or more from project planning to implementation.”

The credit line is load-bearing: Swissgrid reported 100% grid availability in 2025. The operators are not failing; I carry the morning crush on their reliability. The pressure belongs on the structure — the consenting regimes, the procurement rulebooks, the mismatch of clocks — which is where the grid-acceleration proposal before the Council of States this spring points. The infrastructure a generation regrets is rarely the thing that fell; it is the second path quietly deleted from the graph — the spare transformer never ordered, the diversionary line closed to save a budget — until one demand surge arrives and there is no slot left to buy. A grid rebuilt to the same 6,700 km with no redundancy on paper is a single point of failure wearing a public ribbon.

Atelier: The connection application is a design deliverable, not a consultant’s errand, and its date is the project’s date. On anything with a real electrical load — a data centre, a logistics shed with a charging yard, a district heat-pump plant, a rebuilt industrial site — put the grid connection enquiry in the first programme diagram, next to the site survey, not in the technical annex. Your Monday move: on the current live job, send the connection enquiry and ask two questions — when does the connection actually arrive, and what is the design if it arrives four years late? On-site generation, storage, a staged load ramp, a plant room sized for equipment that does not exist yet: the answer changes the building.

Hack: Model the opening date as the slower of two chains, not the sum of the one you manage. Project completion is a max() over parallel chains, not a sum() down your own — which is why shaving the construction programme buys you nothing when the connection is longer.

build   = {"design": 1.0, "permit": 1.5, "construct": 2.0}                    # the clock you manage
connect = {"application": 0.5, "study": 1.0, "transformer": 4.0, "line": 6.0}  # the one you don't
ready = max(sum(build.values()), sum(connect.values()))
print(f"build {sum(build.values())}y · connect {sum(connect.values())}y · opens in {ready}y")

Halve construct and ready does not move. That is the article, executable — and it is the arithmetic the Infrastructure Systems Design school has made for years: a project is a network of coupled constraints solved together, and the honest number lives on the binding one.

Move. Pull up the live job today and find the connection date. If nobody on the team owns the slow clock, that is the job — take it. It is the one date on the project you cannot compress.