Europe Controls Just 5% of AI Compute: Critical Dependency

Europe Controls Just 5% of AI Compute: A Critical Dependency

While the United States monopolizes 75% of global artificial intelligence computing power and China races to close the gap, Europe barely reaches 5%. This figure, drawn from recent analyses of cloud infrastructure and GPUs, paints a picture of strategic dependency that goes far beyond hosting corporate email on Microsoft servers or health data on AWS. It represents the new layer of technological sovereignty.

The alarm is not new, but the geopolitical context sharpens it. The pandemic and tensions in the Taiwan Strait made clear that when supply chains break, each bloc looks after itself. The fact that major AI models—GPT, Claude, Gemini—and the infrastructure to train them are under US or Chinese control is not just a commercial competition issue: it is an access risk conditioned by political decisions.

The Illusion of ASML as a Negotiating Lever

One proposal gaining traction is using ASML—the Dutch company manufacturing lithography machines essential for advanced chips—as bargaining currency. The idea: if Europe holds a monopoly on critical semiconductor technology, it can negotiate better terms for accessing AI chips or even force factory installations on European soil.

However, this argument has a Achilles' heel: ASML is a private company, not an arm of European industrial policy. Nationalizing it or pressuring it to act as a trade weapon would be a move no government has seriously considered, likely triggering a chain reaction. Moreover, as noted in the debate, competitors are already developing alternatives to ASML's EUV lithography. China, in fact, has spent years trying to bypass restrictions and has made pogre.

Energy Bottlenecks and Brain Drain

Another frequently cited obstacle is energy. AI data centers are voracious consumers, and Europe lacks the cheap, abundant electricity of the US—where fracking and natural gas created a buffer—or China's coal and hydroelectric resources. In some European countries, new data center construction has slowed due to grid capacity limits and environmental opposition. Meanwhile, brain drain continues: Demis Hassabis founded DeepMind in London but ended up at Google. European AI talent is trained here but moves to Silicon Valley or big tech firms.

The Mistral Bet and AGI Skepticism

On the positive side, the alliance between NVIDIA and French startup Mistral AI, backed by Macron, has generated some hope. Announced at VivaTech, the deal aims to create open-source models optimized for NVIDIA's Blackwell architecture. Nevertheless, some analysts see it as a patch: Mistral still relies on third-party chips and software, and true sovereignty requires own compute capacity, not just models.

Beyond geopolitics, the debate also questions the hype around Artificial General Intelligence (AGI). It is argued that the promise of conscious AI within three years is pure marketing to attract pension fund investment. Evidence suggests current models do not scale toward consciousness, and singularity—AI's ability to improve itself—is still distant. But even if AGI is smoke, dependency on current infrastructures is very real.

The Illusion of ASML as a Negotiating Lever (Part 2)

Some insist Europe must leverage its position in the semiconductor chain, not to block, but to negotiate agreements including knowledge transfer. The proposal is clear: invest in chip design, advanced packaging, and attract TSMC or Intel with conditional subsidies. In 10 or 15 years, Europe could produce a significant fraction of its own compute. The problem is time and political will: while the EU debates bureaucracy, the US and China advance silently.

---

With these elements, the lingering question is whether Europe can react before the gap becomes insurmountable. Or if, as peine with the internet or mobile phones, the continent settles for being a market for foreign technologies, paying the toll of dependency.