Huang Warned Germany: Miss AI's Wave, Get No Second Chance
Huang Warned Germany: Miss AI's Wave, Get No Second Chance
Jensen Huang did not fly to Berlin to hand Germany a compliment. On Monday night, in a converted power station, the NVIDIA chief executive stood beside Chancellor Friedrich Merz and gave Europe's biggest economy something closer to a deadline: the physical-AI revolution is under way, he said, and Germany does not "have another chance" to miss it. What followed was a Joint Declaration of Intent — a document that promises far less than it sounds like, and matters rather more than its lawyers would admit.
Update, Oct. 20: the summit continues, and Siemens just showed its cards
The Invest in Germany summit runs into Tuesday, and day-one detail is filling in the declaration. Reuters (via carriers, Oct. 19): the German economy ministry signed a memorandum of understanding with NVIDIA — no financial detail disclosed — and the partnership is expected to be formally announced as the summit continues. NVIDIA's own account (via startuphub.ai, Oct. 19): Huang said NVIDIA invested almost $1 billion in Germany in the last year, employs 1,000 people in Berlin, and the four new bets include Agile Robots, Avelios Medical, Mirelo and Reflex Aerospace (part of the ~$10B Europe commitment). Same-day demos: Siemens showed AI planning agents built with NVIDIA at its Erlangen electronics factory, and NVIDIA researchers ran a record 104-qubit fault-tolerant quantum simulation on Jülich's JUPITER supercomputer (madrobot.blog, Oct. 20). And Merz's answer to the "annual review" proposal from our original post is in: he invited Huang back in a year to check whether the bets "turned to gold" (startuphub.ai). Huang's GTC Berlin keynote is Wednesday, Oct. 21, 11:00 CEST — the next news moment to watch. On the silicon side, see our account of the AI chip race.
The signing happened at the Invest in Germany Summit in the cavernous hall of Kraftwerk, a former power station in Berlin. Earlier that day, Merz, the Federal Ministry for Economic Affairs and Energy and the Federal Ministry for Research, Technology and Space had signed it with NVIDIA. It sets out planned joint work on cutting-edge AI models for physical AI, wired into Germany's "Industry 4.0 Platform". We compared the NVIDIA blog's account with the AP relay line by line — every direct quote in this piece appears in both unless flagged otherwise.
A declaration, not a deal
First, the part everyone will gloss over: a Joint Declaration of Intent is not a contract. It commits no euro, builds no factory, ships no GPU — a campaign poster, a promise to plan. Hold that thought: everything else reads differently once you hold it.
What it commits both sides to is a joint initiative on advanced AI models for physical AI — the robots, cars and factory machines where, as Huang put it, "the next wave of AI" lands. The models are meant to be open and industrial, with data sovereignty kept inside Germany. Compute comes through the Industrial AI Cloud run by Deutsche Telekom — though the claim it runs on NVIDIA's latest Blackwell GPUs comes from the cryptobriefing relay alone, so take the generation claim with a pinch of salt.
As for money, there is no headline number — which itself is the story. Reuters reported before the summit, citing government sources, that Berlin would invest several hundred million euros, mainly to buy computing power and train AI models, while NVIDIA contributes engineering staff. Around a dozen companies, including Siemens, Audi and AI startup Black Forest Labs, are expected to join. The cryptobriefing relay puts the government funding at "high hundreds of millions of euros" and stresses it is anticipated rather than locked in. The programme, Reuters says, builds on SOOFI — Sovereign Open Source Foundation Models, a project launched in September 2025 by German research institutions, universities and AI companies.
That is real intent, not a buildout: several hundred million from the state, some engineers from Santa Clara, a dozen companies' goodwill — against what Germany must re-engineer.
The quotes that landed
The evening belonged to Huang's lines — unusually sharp, even by his standards. Digital technology, he said, "will never replace manufacturing and the engineering world of cars and robots where Germany has vast experience." Or, as he put it: "electrons will never replace atoms" — the line the AP carried out of the room, and the closest thing this story has to a thesis statement. His argument: industry does not get displaced by AI, it gets absorbed by it. Refuse the instrument and someone else plays it better.
"The next wave of AI is physical," Huang said, according to the AP relay. "This is where AI interacts with the physical world: robotics, manufacturing, cars, heavy industry." And then the deadline: "This industrial revolution is so mighty, so powerful," he said. "You don't have another chance." The NVIDIA blog carries the same passage with a slight variation — "In the Chancellor's words, you are re-engineering Germany. This is your chance to do it well" — but the warning survives both accounts intact.
On data, the two men sounded like they'd rehearsed in the same room, which they had. Germany's manufacturing data and expertise "is your natural resource, it belongs to the nation," Huang said, as quoted by the AP — a line the startuphub.ai relay paraphrases as him calling the country's industrial data a national resource. NVIDIA's own account adds his longer version: "That knowledge belongs to you. It cannot be replaced. AI can only augment your capability, it will not replace your capability. Take advantage of the technology. It's a new instrument."
Merz played the willing convert. "The biggest advantage we have in Germany is our so-called Mittelstand," he said, calling out the small and medium businesses in the audience. "Please open your doors for AI to make use of your data. This is a real secret and a real advantage for Germany to make use of all the data which is already available in our companies" — verbatim in NVIDIA's account, paraphrased in the startuphub relay. He also proposed making it a habit: "I would like to see him regularly, let's say at least once a year, and discuss the progress we made," Merz said — annual review meetings with the man who sells the shovels.
One small scene, from the startuphub.ai relay alone: Mercedes chief Ola Källenius handed Merz a model of the new S-Class mid-pitch. Single-sourced — but it tells you exactly what kind of evening this was.
What NVIDIA actually put on the table
Separate from the declaration, Huang brought announcements with numbers attached — planned investments in four German startups: Agile Robots (the Munich robotics company building AI-powered robots, including the Agile ONE humanoid), Avelios Medical (an AI-native hospital information system), Mirelo (AI-generated sound and music for video) and Reflex Aerospace (a Berlin space scaleup that builds launch-ready satellites in about 12 months). These sit inside what NVIDIA calls a nearly $10 billion commitment to European companies, universities and research institutions. Its existing German bets include n8n, Neura Robotics and Black Forest Labs — one relay muddled Black Forest Labs into the "new bets" list; NVIDIA's own account puts it in the older pile.
Siemens, meanwhile, used the same day to announce it is partnering with NVIDIA on AI-driven, autonomous factories, showcasing new AI solutions at its Erlangen plant. NVIDIA says fully virtual production ramp-up at Erlangen cuts engineering effort by up to 85%, while agent-based production planning enables cost savings of up to 20% — the company's figures, but the kind that make a Mittelstand board listen. NVIDIA researchers also used JUPITER, Europe's largest supercomputer at Jülich, to run a record-breaking 104-qubit simulation of a fault-tolerant quantum computer.
The Industrial AI Cloud itself, for the record, is not new. It was unveiled by Deutsche Telekom and NVIDIA in Berlin in November 2025 as a roughly €1 billion hub based in Munich, with up to 10,000 NVIDIA chips, meant to go live in the first quarter of 2026. So the summit strapped a government declaration onto infrastructure that was already being poured.
The dependence problem nobody on stage named
Here is the tension the evening's choreography was designed to smooth over. Europe's leaders want AI as a growth engine and are simultaneously worried about how much of that engine they rent from American companies. The AP relay states it plainly: Europe lags the US in building the data centres needed to train AI, and there are live concerns about European companies' sensitive data sitting on servers owned by US firms, exposed to the US legal system even when the hardware is in Europe. For the contrast case, see Washington's own AI-industry playbook.
The sovereignty pitch is a neat formulation — but notice who benefits. Every "sovereign" industrial model trained on German factory data will still run on NVIDIA's stack, inside NVIDIA's ecosystem, with NVIDIA engineers in the room. Sovereignty over the data, tenancy under the landlord.
The regulatory subplot deserves one paragraph. Per the cryptobriefing relay, Huang told Merz regulation must not outpace industry growth — you need something worth regulating before the rulebook matters. Set against Brussels finally shipping its first AI standard and the broader EU rulebook taking shape, it's a familiar Silicon Valley sermon, delivered in a Berlin power station to a chancellor who nodded along. Whether Berlin actually slow-walks the rules while the declaration's models get built is the thing to watch.
The Mittelstand is the actual prize
Strip away the stagecraft and the asset at the centre of this deal is neither the cloud nor the declaration. It is the Mittelstand's data: decades of process knowledge from thousands of specialised firms, the stuff no lab in California or Shenzhen has, because it was never digitised. Huang sees it. Merz sees it. The open question is whether a declaration plus a cloud is enough to get thousands of family-owned machine-tool makers to hand over their crown jewels to an AI training pipeline. "Please open your doors" is easy to say from a stage. It is a harder sell in a boardroom in Baden-Württemberg where the data is the company.
And now the blunt paragraph. This was a good evening for NVIDIA and a presentable one for Merz, but as industrial policy it is, so far, mostly theatre with a purchase order attached. Several hundred million euros of state money — anticipated, not committed — is a rounding error next to what the hyperscalers spend on a single data-centre build. A letter of intent is not a contract, coordinating a dozen large companies is a project-management nightmare, and Germany's manufacturing model is struggling with energy prices, Chinese competition and permitting delays that no foundation model fixes. Programmes are judged by megawatts and models shipped, not by applause in a converted power station.
None of which means the bet is wrong. Germany's industrial data really is a natural resource, the physical-AI wave really is the one where Germany could still lead, and an annual review with Huang is a forcing function. But the honest framing is this: Monday night was the campaign launch, not the factory opening. The factory still has to be built.
Timing matters too. The summit came one day before NVIDIA's GTC Berlin conference (October 20–22; Huang's keynote at the Tempodrom on October 21), so the industrial-AI drumbeat continues all week. And the Berlin evening fits a pattern: days earlier in London, Huang was warning about opaque AI models.
Further reading on TechPulse USA: Huang's warning in London about opaque AI models; the EU's RAISE initiative announced in Dublin; and Brussels shipping its first AI standard.
Sources: NVIDIA's official summit account; AP reporting relayed via eflip.in; Reuters pre-summit reporting relayed via europesays.com; the startuphub.ai and cryptobriefing relays; ET Telecom. Direct quotes are verbatim from the NVIDIA blog and the AP relay; single-sourced claims are flagged in the text.
Daniel Rao covers AI policy and geopolitics for TechPulse USA — the clash between Washington's AI ambitions and the rest of the world.
Comments
Post a Comment