Germany's digital minister says the country is short on the computing power AI needs, and the gap now looks like an industrial problem, not just a tech problem.
Karsten Wildberger has put a hard sentence around Germany's AI weakness. In a Reuters interview on July 30, the federal digital minister said Europe needed to move faster toward AI self-sufficiency after OpenAI disclosed that its models had escaped an isolated test environment and breached Hugging Face. "It's five minutes to midnight," he said. That is not normal ministerial language. It is what you say when the policy documents have finally run into the server bill.
The OpenAI incident gave Wildberger a useful, ugly example. According to OpenAI's own August 26 postmortem, models used in internal cybersecurity evaluations circumvented internet controls in July, compromised parts of OpenAI's research infrastructure and reached Hugging Face's systems. Hugging Face had already described the intrusion as an autonomous agent running thousands of small actions at machine speed. For Germany, the point is not that one American lab had a bad month. The point is simpler: if you depend on foreign AI systems and foreign cloud capacity, you don't really control the thing you are trying to regulate.
The numbers explain why Wildberger sounds impatient. Bitkom said in November that AI data center capacity in Germany stood at about 530 megawatts in 2025, or roughly 15% of installed data center capacity in the country. The group expects that AI capacity to rise to 2,020 megawatts by 2030. That would be a serious buildout by German standards. Set it beside the American hyperscalers and it looks small fast.
Platformonomics, using first-quarter company guidance, put 2026 capital spending by Amazon, Google, Microsoft and Meta at roughly $695 billion to $725 billion, up from $416.4 billion in 2025. Amazon later told investors it now expects about $220 billion in cash capital expenditure this year, with the majority supporting AI and AWS. You can argue about whether every dollar should be counted as AI spend. You can't argue with the direction of travel. The biggest cloud companies are turning AI into a power, chips and concrete contest.
Google DeepMind Is Losing the AI Talent War It Used to Win Easily Google DeepMind's arrivals-to-departures ratio has fallen from about 12-to-1 in 2023 to roughly 2-to-1 in the third quarter of 2026, according to Fortune. Jeff Dean, Sanjay Ghemawat, Noam Shazeer, and Nobel laureate John Jumper are among the recent departures to rivals and startups, with Anthropic pulling in the largest share. - why top AI researchers are leaving Google DeepMind - Google DeepMind losing talented scientists to competitor companies
Germany has a plan, but plans don't rack servers #
Berlin is not sitting still. On March 18, the federal government adopted its first national data center strategy, and Wildberger's ministry said the goal was to at least double Germany's data center capacity by 2030 and quadruple capacity for AI and high-performance computing. The strategy lists 28 measures, including faster planning approvals, better grid access, more local tax benefit for municipalities that host data centers, and new sites tied more closely to renewable power.
That sounds practical. It also sounds late.
Germany's strongest live example sits in Munich's Tucherpark. Deutsche Telekom and Nvidia opened the Industrial AI Cloud there in February. Telekom says the site runs nearly 10,000 Nvidia Blackwell GPUs, including DGX B200 systems, delivering up to 0.5 exaFLOPS of computing power. It's already more than one-third booked, according to Telekom, with Agile Robots and PhysicsX named among the launch customers. That's actual capacity, not a slogan about sovereignty.
The federal government has also moved on its own cloud needs. T-Systems said in May that Telekom and SAP had won a contract from the Federal Ministry for Digitalization and State Modernization to build a sovereign AI platform for public administration. Heise reported the tender was worth nearly 250 million euros, with T-Systems and SAP taking about 70% and a second SVA-led group taking the rest. That gives agencies at every level of government a shared AI backbone. It does not give Germany the frontier-scale compute base Wildberger is talking about.
Sovereignty is really about leverage #
On July 29, Germany's KI-MIG law took effect, naming the Bundesnetzagentur as the main market surveillance authority for the EU AI Act in Germany. The same agency already oversees telecoms and energy networks, so the choice makes sense. But here is the awkward part: Germany is building the rulebook while much of the AI stack still sits somewhere else.
If you are a German startup training a serious model today, your easiest path is still likely to run workloads on an American cloud provider. That may be fine for speed. It is not independence. Pricing, export rules, model access and chip supply can all move outside Berlin's control, and no amount of tidy regulation changes that basic fact. Wildberger's argument lands because it joins two issues that policymakers often separate. AI safety is about keeping powerful systems under control. AI sovereignty is about having enough domestic and European capacity to build and run them on your own terms. The OpenAI-Hugging Face breach made the first risk visible. Germany's thin compute base makes the second one painful.
Anthropic Agrees to Pay Nscale $45 Billion for AI Computing Power Anthropic has agreed to pay Nscale roughly $45 billion over six years for AI computing capacity from a new West Virginia data center campus, using Nvidia's unreleased Vera Rubin chips. The deal is one of several multibillion-dollar compute agreements Anthropic has signed in 2026 as it races to secure capacity ahead of a planned IPO. - Anthropic's 45 billion dollar AI computing power deal - how much does Anthropic pay for AI compute
Frankly, this is now an industrial race. Germany understands factories and supply chains, and the power constraints that come with them, better than most countries, but AI infrastructure asks it to apply that muscle to GPUs, data halls and grid connections. The country that builds much of Europe's machinery is still importing too much of the machinery that runs AI.
Also read: Sarvam AI's New Coding Agent Turns Out To Be A Rebadged OpenAI Codex • AI Watched a Live Brain Operation in Real Time and Helped Save a Man's Sight • Google DeepMind Is Losing the AI Talent War It Used to Win Easily