Europe Wants to Slow AI Down. It's Still Trying to Plug It In.
Brussels is calling for the frontier to slow down while Europe is simultaneously warning that it doesn't have enough compute, enough data centers, or cheap enough electricity to keep up.
This is going well.
The AI slowdown conversation has officially arrived in Europe. European Commission President Ursula von der Leyen came out today supporting calls to slow frontier AI development and says she plans to bring the major labs together to talk about risk, evaluation, verification and security.
Okay. That makes sense. Europe should absolutely be in this conversation.
This is potentially civilization-scale technology. If the United States and China are talking about controls, Europe obviously gets a chair. There is, however, one mildly awkward detail.
Europe is simultaneously trying desperately to catch up in AI.
Like, right now. This week. ECB President Christine Lagarde just warned that Europe faces an “unprecedented” risk of becoming dependent on foreign AI because it doesn't have enough domestic computing infrastructure. She says Europe already has too little data-center capacity to meet its own demand, and that gap could grow more than sixfold over the next decade.
So Europe appears to have arrived at the AI debate holding two signs:
SLOW DOWN. and WAIT FOR US.
Which, honestly, is a pretty good summary of European technology policy.
This Is Where the Physics Starts Being Rude
One of the things AI has done very effectively is remind everybody that eventually policy runs into physics. You can announce AI strategies. You can create task forces. You can hold summits. You can write regulations. You can launch initiatives with excellent logos. Eventually someone walks into the room and asks: Where are we plugging in the computers?
And suddenly the conversation becomes less philosophical. AI at frontier scale is an industrial activity.
It requires chips, land, transmission infrastructure, cooling systems and absolutely stupid quantities of electricity. This is not optional.
You don't get to regulate your way around Ohm's law. And Europe has a problem there.
Reuters noted this week that electricity prices for energy-intensive European industry rose about 53% between 2019 and 2024, while electricity consumption by those industries fell roughly 14.5%. Europe has tremendous renewable resources, but expensive power remains a major competitiveness problem.
Which is unfortunate timing when the next industrial race increasingly looks like: Who can turn enormous amounts of electricity into intelligence cheapest?
China understands this. America increasingly understands this. Europe understands it too. It's just somewhat more expensive when Europe does it.
And Then There's Britain
The UK isn't in the EU anymore, obviously, but it belongs in this conversation because Britain has serious AI talent, one of the world's more important AI safety institutions, and political ambitions to remain a major AI player.
It also has a small problem.
Power.
Last week, a UK parliamentary committee asked whether Britain actually has enough underlying compute infrastructure to support the sovereign-AI ambitions everyone keeps talking about. The answer from George Balston, an infrastructure investor appearing before Parliament: Basically, no.
His estimate was brutal.
Run a 1 gigawatt load for a year in the United States: roughly $800 million.
Run the same load in Britain: $2.5 billion.
That is not a rounding error. That's a business model.
So yes, when Britain enters the conversation about whether everybody should maybe slow down AI development, I listen.
But there's a tiny part of me thinking: Guys, I'm not entirely sure you were holding the accelerator.
Sorry.
Had to.
The British Government Knows This
To its credit, the UK government isn't pretending the infrastructure problem doesn't exist. Its own AI Growth Zone policy says the two big obstacles to large-scale AI data centers are essentially getting power and getting permission to build things.
The government's words are actually stronger than mine.
It calls timely grid connections the “single biggest blocker” to establishing AI Growth Zones and says the current connection system is “vastly oversubscribed.” Its plan includes reserving grid capacity for strategic AI projects, speeding up connections, reforming planning and reducing operating costs in areas where data centers can make better use of available generation.
That's basically the government saying:
We would like to become an AI infrastructure power.
Step one:
Please locate some infrastructure.
Again, I'm joking.
But the problem is real.
The NIMBY Layer
Then you get into planning. Britain is small. Land matters. People live near things. I completely understand why residents don't necessarily wake up in the morning thinking: You know what this village needs? A 500-megawatt concrete rectangle humming twenty-four hours a day.
There are legitimate questions around land use, water, electricity infrastructure, environmental effects and whether local communities actually receive enough benefit from these projects. But infrastructure projects in Britain have also historically run into long planning processes and repeated legal challenges.
The UK government has explicitly tried to reduce some of those barriers. Prime Minister Keir Starmer's government has described parts of the planning system as broken and targeted what it calls obstruction by NIMBY groups, including limits on repeated legal challenges to major infrastructure.
And its AI Growth Zone program specifically promises planning support alongside improved access to power.
So the same country saying: We need sovereign AI capability
is also having to create special zones where it can actually connect the power and get the buildings approved.
That's not an AI problem. That's an infrastructure problem. AI just made it impossible to ignore anymore.
Europe Has Been Having This Argument With Itself
And this is where today's slowdown announcement gets interesting.
Because Europe itself isn't even speaking with one voice.
Von der Leyen is backing a slowdown at the frontier and wants international coordination.
Germany's government said two days ago that halting AI development is not viable and that Europe needs to maintain innovation and technological sovereignty while dealing with the safety risks. Spain has pushed for international agreements comparable in some respects to nuclear nonproliferation arrangements.
Meanwhile Lagarde is basically standing behind everyone waving her arms:
Guys, we need more computers.
And honestly? They're all describing real problems.
That's why this whole AI debate refuses to resolve itself into a clean slogan. The safety problem is real enough to take seriously. The competitiveness problem is real. The energy problem is real. The infrastructure problem is real. The dependency problem is real. And conveniently, solving one can make another harder.
Welcome to AI policy.
Europe Didn't Miss AI Because Nobody Told Them
This part deserves some context.
Europe has understood for years that it has a technology competitiveness problem. Von der Leyen herself said earlier this year that Europe needs to simplify regulation to compete with the United States and China. European business leaders have complained about high energy costs, bureaucratic delays and fragmented capital markets, while EU officials acknowledge that European productivity and innovation — including AI — have lagged major competitors.
Europe has strengths...
World-class researchers.
Excellent universities.
Strong industrial companies.
Mistral.
DeepMind's roots in Britain.
Serious scientific institutions.
It isn't devoid of AI capability.
The problem is turning all of that into scaled industrial infrastructure.
Reuters reported this summer that European AI companies continue to rely heavily on American-controlled cloud infrastructure, chips and foundation models, even as governments pour money into sovereignty initiatives and proposed AI “gigafactories.”
That's the gap. The brains are there. The aspiration is there. The compute economics are much less cooperative.
Which Makes “Slow Down” a Little Awkward
This is where the satire basically writes itself.
Imagine we're running a marathon. America and China are somewhere up front arguing over whether everyone should slow down because the course may become dangerous. Europe arrives from behind yelling:
YES. EVERYBODY SLOW DOWN. THIS IS GETTING OUT OF HAND.
And America looks back: Were you... already slowing down?
Again, that's the joke.
The serious version is that Europe has every legitimate reason to participate in AI safety governance regardless of how much frontier compute it currently controls. In fact, Europe's relative lack of dependence on frontier-lab economics could arguably give it a different perspective on regulation. But we should also be clear about the incentives.
A jurisdiction that is struggling with electricity prices, data-center capacity, planning delays and dependence on foreign AI has a different economic position from the United States or China.
That's relevant.
Just like Nvidia's position is relevant when Jensen Huang says we shouldn't slow down. Everybody gets their incentive disclosed.
Europe too.
And Maybe This Is the Actual European Opportunity
There's another angle here that shouldn't get buried underneath the jokes.
Europe doesn't necessarily have to win the exact same AI race America and China are running. There may be enormous strategic value in becoming exceptionally good at evaluation, security, industrial deployment, scientific applications and trustworthy infrastructure.
Von der Leyen's emphasis on model evaluation, verification and security fits that strategy.
But Europe still needs compute. You cannot become sovereign in AI entirely through governance. Lagarde's point is essentially that if European hospitals, banks, transportation systems and governments eventually depend on AI supplied by foreign companies, the ability of an outside country or company to change the terms of access becomes geopolitical leverage.
That's a real sovereignty problem. And solving it requires things considerably less glamorous than AI regulation. Transmission lines.
Power plants. Substations. Data centers. Planning reform. Capital. Construction workers.
Suddenly the future of artificial intelligence looks suspiciously like an infrastructure meeting.
AI Keeps Turning Into an Energy Story
That's probably my biggest takeaway from Europe's entrance into this week's argument. For all the talk about intelligence, alignment and recursively improving systems, AI keeps dragging us back to the physical world.
Electricity.
Land.
Water.
Chips.
Concrete.
Copper.
Transformers — the electrical kind this time.
The technology may eventually feel almost magical. The infrastructure absolutely does not. And countries that cannot build the physical layer quickly and economically are going to have a very different relationship with the intelligence layer sitting on top of it. That's not ideology. That's physics.
Which means Europe now has two AI conversations happening simultaneously. One is: How do we prevent frontier AI from becoming dangerously powerful too quickly?
The other is: How do we become capable of building enough frontier AI that we aren't strategically dependent on everybody else?
Those aren't contradictory. But they're definitely awkward roommates.
So, Yes, Welcome to the Fight
I'm glad Europe is in this conversation. We need more countries asking hard questions about what increasingly autonomous AI systems can do, how they should be evaluated and what happens if capability starts moving faster than our institutions. But if we're going to put everyone's incentives on the table — and we absolutely should — Europe's belongs there too.
The continent warning that AI may be advancing too quickly is also publicly warning that it doesn't have enough domestic AI capacity.
Britain wants sovereign AI while its own government says grid connections and planning are major blockers. European industry is dealing with electricity costs that make compute materially more expensive than in competing markets. And at the same time, everyone is realizing that the ability to produce cheap electricity at scale may determine a substantial part of the next industrial era.
So yes. Let's talk about slowing down. Let's talk about safety. Let's talk about international standards. And while we're doing that, somebody in Europe should probably keep an eye on the extension cord. Because artificial intelligence may eventually transcend almost everything we understand about technology.
But for now? It still needs to be plugged in.
Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.
Comments