Germany currently has about 530 MW of AI data center capacity. Foreign companies operate most of it. That detail sits inside a press release about Europe's infrastructure boom, and almost no one is quoting it.
The headline number from the European Data Centers Association is €176 billion in investment from 2026 to 2031. That is double the prior estimate. Commission EVP Henna Virkkunen wants EU capacity to triple in five to seven years. Those are real commitments on paper. The problem is who is actually deploying the capital: Google put €5.5 billion into Germany through 2029. AWS has spent $230 billion in Europe over 20 years and added a $7.8 billion Sovereign Cloud push. Microsoft, Amazon, and Meta are all building. The EUDCA's own Winterson said, without much fanfare, that American players represent "by far the largest share" of that €176 billion.
This is not investment in European AI infrastructure. It is American companies building their own infrastructure on European soil, under European grid connections, occasionally with European tax incentives. Calling that a European AI buildout is like counting every McDonald's franchise as a local restaurant win.
One Bet That Actually Counts
Polarise is the exception worth watching. The company is building a 30 MW AI data center in Amberg, backed by €117 million from Macquarie, targeting mid-2027, and explicitly refusing to rely on subsidies. The plan is to prove the commercial model works against foreign-controlled capacity, powered by renewables, without a government backstop. If it works, it is a blueprint. If it does not, the failure will demonstrate that European risk capital still cannot match what American hyperscalers absorb without blinking.
For comparison: Jensen Huang put global AI infrastructure capex at $700 billion this year, mostly Alphabet, Amazon, Meta, and Microsoft, mostly in Virginia, Georgia, and Pennsylvania. McKinsey is forecasting $6.7 trillion in cumulative data center investment by 2030. Polarise's 30 MW is a proof of concept, not a counterweight.
The honest tension in my argument is this: American investment does fill a real gap. If AWS and Google pulled out tomorrow, European builders could not replace that capacity in any useful timeframe. Winterson acknowledged it directly: their withdrawal would leave a "considerable gap." That is true. It does not make dependency a strategy.
What European Builders Actually Need to Do
The Deutsche Telekom and Nvidia deal in Munich, €1 billion, roughly 10,000 Blackwell GPUs, opened in 2026 and looks impressive. But Nvidia sells the hardware and Deutsche Telekom runs the pipes. The compute architecture, the supply chain, the pricing leverage all originate in California and Santa Clara. Europe has the building; America owns the engine.
European founders and infrastructure operators should stop citing American hyperscaler investment as evidence of ecosystem health. It is a dependency metric, not a growth metric. The target should be replicating the Polarise model at scale: local capital, commercial viability without subsidies, renewable power, and ownership that stays inside the jurisdiction.
Tripling EU data center capacity in five years is achievable. Doing it while ensuring European builders own the resulting infrastructure is a different and harder problem. The €176 billion figure means very little if the capacity belongs to Microsoft's Azure or Google Cloud when the contracts expire.
Europe is renting its own AI future. Polarise is one early sign that someone noticed.
