Supercharging European AI in a non-superboring way; UAE and KSA ramp up AI investments; the age of "manager nerds" is upon us; a fraction of AI experiments are delivering returns, CEOs say

Elon Musk's power play at the US Copyright Office; AI research takes a backseat to product innovation; inside the race for legal AI; Big Tech aims to shut down California's AI rules

May 16, 2025

∙ Paid

Nothing gives my European heart more palpitations than the idea of listening to EU politicians pontificate upon the question “Can Europe become an AI continent?” when they just passed legislation in the form of the EU AI Act that pretty much ensures the answer is a hard no.

Yet, I somehow found myself this week livestreaming Politico’s AI & Tech summit in Bruxelles (yes, I will forever refuse to spell it Brussels). The most interesting part of the event by far was the Supercharging European AI panel where the conversation focused on a topic I’ve written about previously - the European AI gigafactories. The star of the debate was Bulgarian MEP Eva Maydell who laid out the real challenges facing Europe in building out AI infrastructure: there’s a very visible lack of a coherent strategy at the EU level on how and where the bloc should compete.

I do not believe that a strategy is one where you just announce a big amount of funding without knowing for what purpose. And this is where, in a certain manner, we try to copy others. And I think we need to have our own, unique way. So a bit more of a vision and strategy […] that is backed by data and necessity. And not backed by the fact that others are investing big amounts [of money]. We should do the same, because our strengths are different. - Eva Maydell

Leave it to the Eastern European to deliver the cold, hard facts.

Okay, here are some numbers that put in perspective what Eva is saying. The European AI gigafactories aim to house about 100,000 GPUs each which is less than half of what is inside the Colossus data center built by xAI in Memphis. These gigafactories will take years to build while xAI did theirs in 122 days. Hyperscalers like Meta plan to build a 2GW data center with 1.3 million GPUs and Microsoft will likely buy close to two million GPUs this year. The reality is that there are only three nations that can come close to the scale of AI infrastructure being built in the United States:

China, a country currently locked out of high performance NVIDIA hardware due to export restrictions set by the US
Saudi Arabia announced Humain, an AI subsidiary of the Saudi Public Investment Fund, and committed to building 500MW data centers over a five year period
The UAE announced a 5GW AI Campus to be built by G42 in partnership with American technology companies

So what should Europe do? According to Eva, it should “be strategic and operate like a surgeon” as “it would be great if [Europe] can identify the few big ideas that could propel our economic and industrial base.”

Here are three interventions the EU can make to keep the patient from going into cardiac arrest.

First off, green computing. The American grid is a mess so hyperscalers are currently scrambling to use gas turbines or revive closed down nuclear or coal power plants. Europe’s highly interconnected electricity grid ensures a more stable electricity supply and minimizes the duration of blackouts. According to the system average interruption duration index, blackouts last on average 0.9 hours in Europe, while reaching more than 10 hours in the US. France in particular has done an amazing job of building out their grid powered by nuclear reactors. The rest of Europe could double down on energy abundance and reliability, and build more nuclear, solar, wind and hydro-based power plants.

Secondly, encouraging the market to adopt AI faster. European companies have been incredibly slow at adopting AI compared to their American and Chinese counterparts: only 13% of businesses adopted AI tools in 2024. That is a shockingly low number. The EU Commission has been talking a lot about simplifying and harmonizing regulation but, in my view, that in itself will not solve this challenge. Nor will the artificial promotion of “EU champions.” Instead, adoption comes when you lead by example and empower those around you to feel safe in experimenting. In other words, you need to deploy fast and learn as you go. For example, DG CONNECT could work with regulators from EU member state to build a program where every few months they speak with the large companies to check in how they’re building or deploying AI, and publicly celebrate their success. That could create a cultural shift away from EU-as-a-regulator to EU-as-a-trusted-adopter.

Finally, launch mission-oriented vertical moonshots that motivate scientists and researchers to stay in Europe and push the frontier of AI. Auto-regressive models have taken us very far but they are clearly reaching a scaling plateau. The answer to what comes next should be uncovered in European universities. Similarly, all AI training happens on GPUs but what if there’s an architecture that’s 1000x times faster or cheaper? To make the deal even sweeter, you can scale the moonshot program from the foundation or infrastructure layers all the way to applications such as industrial robots or chemical processes (drug discovery, in particular).

Focusing on these three edges converts Europe’s strengths (norm-setting power, engineering for efficiency, and unique domain assets) into a competitive equation that is not decided solely by the size of their data centers.

And now here are the week’s news.

(But before you scroll away, I’m hiring a corporate affairs manager for my team.)

❤️Computer loves

Our top news picks for the week - your essential reading from the world of AI

Keep reading with a 7-day free trial

Subscribe to Computerspeak by Alexandru Voica to keep reading this post and get 7 days of free access to the full post archives.