Sovereignty Has a Power Bill
The part where Europe gets the invoice
In April 2025, the European Commission announced it would mobilise €200 billion to make Europe an “AI continent” — €20 billion of it earmarked for up to five AI gigafactories, each a cathedral of a hundred thousand processors.
The phrase “AI continent” might make our American friends smile — the indulgent smile you give your grandpa when he announces he’s finally getting on Facebook. It’s a big aspiration from a place that doesn’t yet own a single layer of the stack. But the money is real, the ambition is serious, and you wouldn’t be wrong to read it as Europe finally getting its shit together. Or at least trying to.
As always, there’s another angle to it.
That same Europe needs €584 billion of grid investment by 2030 just to keep the lights on as electricity demand climbs — a bill it has not figured out how to pay. We are writing cheques in Brussels for the compute, and quietly hoping the wires magically show up on their own.
They will not. Well, not magically at least.
The queue outlasts the model
In the last article I argued that sovereignty isn’t blocked by software, or even by chips. It’s blocked by electrons. Let me show you exactly where the block lives.
As I’ve mentioned — a single modern AI cluster draws as much electricity as a quarter-million homes. Anthropic expects that developing a single frontier model will need a 5-gigawatt data center by 2028 (equivalent to the output of roughly five nuclear reactors). In 2026, that’s roughly 6% of Poland’s entire installed generating capacity (75 gigawatts), ~4% of the UK’s (about 115), and ~3% of France’s (roughly 156).
A hyperscale data center can be built in eighteen to twenty-four months. Getting it connected to the grid in one of Europe’s five big data-center hubs — Frankfurt, London, Amsterdam, Paris, Dublin — takes seven to ten years. In some cases thirteen. At the same time a frontier model is superseded in twelve to eighteen months (Opus 3.x released in Mar 4, 2024, and Opus 4.x first released in May 22, 2025).
As of January 2025, more than 11,900 businesses were queued for a grid connection in the Netherlands alone, and operators have started openly rationing power. West London has connection quotes stretching into the 2030s, with the transmission upgrades that would relieve it not due until 2037.
And this isn’t for lack of a plan. The €584 billion figure comes from the EU’s 2023 Action Plan for Grids — a document that correctly diagnosed every problem above. Two and a half years on, the verdict is in: the ambition went up, but the wires didn’t. Roughly 1,700 gigawatts of renewable and hybrid projects now sit in connection queues across 16 countries — more than triple what’s needed for 2030, and about double the EU’s entire installed renewable fleet. In 2024 alone, €7.2 billion of clean power was thrown away across just seven countries because the grid couldn’t carry it. The plan’s voluntary tools were so clearly not enough that in December 2025 Brussels had to come back with a binding Grids Package — and a price tag that had ballooned to €1.2 trillion by 2040.
The chip was a chokepoint US controlled. The wire is a chokepoint Europe built for itself.
Ireland’s example
If you want to see where this ends, look at Ireland. Ireland did everything right by the old playbook. It courted every hyperscaler on Earth, built the Silicon Docks, became the data-center capital of Europe.
It had consequences, though.
In 2024, Irish data centers consumed 22% of the country’s entire electricity — up from 5% a decade earlier, and more than every urban home in the country combined. The IEA says no other country carries a heavier load, and forecasts put it near a third within a few years.
So Ireland froze new connections. And when it finally lifted the freeze at the end of 2025, it replaced it with a sharp solution: large new data centers now have to bring their own dispatchable power — onsite generation or storage matching what they draw. In short — build your own electrons, or don’t build at all.
I don’t think that’s an Irish quirk. It’s more of a preview.
But renewables!
Yeah, yeah, yeah. Europe is crushing it on clean power. I know. And it’s true — in 2024, renewables hit 47.5% of EU gross electricity consumption, up from 15.9% in 2004. By 2025, solar and wind were outgenerating fossil fuels altogether, with over 70% of EU electricity coming from low-carbon sources. Genuinely a triumph.
Hold the applause for one paragraph.
The headline hides the gap. Take my own country. Poland crossed 50% renewables by installed capacity in 2025 — a great press release. By electricity actually generated in 2024, renewables were about 29%, and coal still supplied 57%, the highest share in the EU (which won’t surprise anyone who knows the country’s history, where coal was less a fuel than a national institution, the same industrial heartlands that, alongside the Gdańsk shipyards, gave us Solidarność). The distance between what you’ve installed and what you can dispatch at 3 a.m. is the whole problem in one statistic.
A renewable grid is optimised for the average megawatt. An AI cluster needs the instantaneous gigawatt — at three in the morning, in February, with the wind down and the panels dark. And the two failure modes are different. A long training run that loses power doesn’t dim the lights; it loses everything since the last checkpoint. Serving the model to users is the opposite problem: it has to be on essentially always — what data centers are built around. Alas, this is neither how the wind nor the sun works.
Which is why the most striking line in European energy policy last year came not from a utility but from Ursula von der Leyen, who told the 2025 State of the Union that Europe needs “more homegrown renewables — with nuclear as a baseload.“
And batteries don’t close the gap. Europe’s fleet is growing fast — 40 GW by the end of 2025 — but today’s batteries average just 1.5 to 2 hours of duration. They’re brilliant at shifting cheap midday solar into the evening peak, and no help on the third still, grey day of a February high-pressure system. Storage makes a renewable grid smarter; it doesn’t make it always-on. For that you still need firm capacity underneath — long-duration storage, and the dreaded N-word, nuclear.
Read von der Leyen’s line, then, as the Commission quietly conceding that “just build more wind” is not, on its own, an AI energy strategy. You also need the boring, dispatchable, always-on megawatts — and that’s exactly where Europe’s clock runs slowest.
France’s first new EPR2 reactor won’t switch on until 2038. Britain’s Sizewell C took its final investment decision in 2025 and faces nine-to-twelve years of construction. Poland’s first reactor: somewhere between 2036 and 2040.
For comparison, the United States is bringing the mothballed Three Mile Island reactor back online by 2028 under a Microsoft power deal — about four years from announcement to switch-on. America is restarting old reactors faster than Europe can build new ones.
The race
Let’s get a little uncomfortable by seeing where we are in comparison.
In 2024, China added 429 gigawatts of new generating capacity in a single year — a 21% jump, the most any country has ever added. Of that, 356 GW was wind and solar: roughly four and a half times the EU’s additions, and close to the entire installed wind-and-solar base of the United States. It also waved through 54 GW of new fossil capacity, because when you’re building baseload you take it where you can get it.
On the other side of the race, the American private sector. OpenAI’s Stargate project planned nearly 7 GW of AI capacity across new sites in nine months, more than $400 billion committed, ahead of schedule. To put that 7 GW in European terms: it’s larger than the entire peak electricity demand of Ireland, which peaks at roughly 5–6 GW. One company, one project, secured more power for AI than a mid-sized European country uses for everything.
Europe’s answer, the gigafactory programme, won’t break ground until 2027.
The €200 billion compute cheque and the €584 billion grid gap aren’t two separate stories. They’re the same story. The first is what Europe wants. The second is what Europe is standing on. If you ask me, right now the foot is going through the floor.
The bypass?
Now, the industry is not waiting around for the wire. It never does. The good thing about capitalism is that wherever there’s money to be made, someone will find a way.
If the public grid quotes you a decade, you build your own. Across Europe, data-center operators are putting up gas turbines, co-locating with dedicated wind and solar, turning their backup batteries into grid assets — anything to skip the queue. One estimate puts the AI-dedicated data-center buildout in Europe at around €176 billion through 2031, much of it now routed around the public grid. In the US, a single turbine-maker’s order book has hit 100 GW, sold out past 2028.
Clever. And for any individual operator, completely rational.
But notice what just happened. We started the last article worried that a foreign government could flip a switch on our models. The escape route depends entirely on who owns the turbine. A European operator powering a European data center with European gas? That’s real sovereignty — exactly the point. But increasingly the builder and the owner are the same American hyperscaler. In that case the private plant doesn’t shorten anyone else’s queue, doesn’t relieve the national grid, and answers to no European energy plan. It’s dimensioned for one data center and switchable by one owner — and that owner isn’t European. You haven’t escaped the foreign switch. You’ve poured concrete around it.
The one chokepoint you can actually build
So here’s where I land, and it’s less bleak than it may sound.
Of all the layers in the AI stack, the grid is the only one Europe can fully own. The lithography machines are Dutch but singular (and US controlled). The best open weights are American. The chips are American and Taiwanese. But a transmission line through Lower Silesia, a battery park in Iberia, a reactor on the Baltic coast — those can be ours, end to end, no foreign letter required to keep them running.
Can be — with one asterisk. A grid wired with Chinese transformers and battery cells just swaps dependence on Nvidia for dependence on someone else. The rule is the same one we keep relearning: own the layer and its supply chain. The difference is that here, unusually, Europe isn’t starting from behind — in transformers, HVDC links and cabling it has real champions (Siemens Energy, Nexans, Prysmian). It has no such thing for frontier chips or models.
That’s the inversion worth sitting with. The thing everyone treats as the boring, capital-intensive afterthought — the wires, the firm megawatts, the storage — is the one rung of the ladder no one can switch off from abroad. Europe spent a decade chasing model sovereignty and chip sovereignty, the two things it can least control. The sovereignty it can build was sitting in the substation the whole time.
It is unglamorous. It is a permitting fight, a financing fight, a fight with everyone who’d rather not have a pylon in their sightline. €584 billion of fight. But it’s a fight with a known answer — pour the concrete, string the wire, sign the nuclear offtake, stop treating the grid as plumbing and start treating it as the strategic asset it has quietly become.
So it’s not so much that sovereignty is blocked by electrons. It’s not blocked. It’s just unbuilt.
And unbuilt is the one problem on this list that money and a decade of nerve can actually solve. The question isn’t whether Europe can build the grid. It’s whether it starts pouring concrete now — or finishes paying the power bill in someone else’s currency.