Germany and Nvidia sign a pact to build AI for factories and robots

Germany has made its bet on artificial intelligence official, and the partner is Nvidia. At the first Invest in Germany Summit in Berlin on Monday, 19 October 2026, Chancellor Friedrich Merz, the Federal Ministry for Economic Affairs and Energy and the Federal Ministry for Research, Technology and Space signed a Joint Declaration of Intent with the US chip company. The goal, in the government's own words, is "to systematically bring industrial AI into widespread use and to establish a state-of-the-art industrial AI ecosystem in Europe."

The centre of the agreement is a joint initiative to develop advanced AI models for what the industry calls physical AI: models that understand and act in the real world, the kind that steer robots, run factory lines and train autonomous vehicles. Nvidia also said it will make its AI models available not only to large industrial groups but also to small and medium sized companies and research institutes. That evening, Merz and Nvidia founder and CEO Jensen Huang sat on stage together at Kraftwerk, a former power station in Berlin, and spent most of their conversation on one theme: German engineering plus AI.

This article is based on Nvidia's own account of the event, the federal government statement as reported by tagesschau, Handelsblatt, which has seen the declaration, and the Siemens press release published the same day.

What was actually signed

A declaration of intent is a political document, not a contract with a price tag. It records what both sides want to do and who is responsible, and it opens the door to concrete projects later. According to Handelsblatt, which has read the paper, the two ministries will lead the work on the government side. The document states that the signatories are convinced AI is "a fundamental requirement" for the long term competitiveness, productivity growth and innovative strength of German industry.

The initiative will build on the Plattform Industrie 4.0, an existing network of politics, industry, research and trade associations that the paper calls the central strategy and delivery network for industrial AI in Germany. A task force with three focus areas is planned. Handelsblatt had first reported the plans on Thursday, 15 October, and described two main fields: physical AI for training robots and autonomous vehicles, and powerful models that can simulate single production steps or entire factory processes.

What the declaration does not contain is just as important. The government gave no figure for public money. tagesschau cites a media report that Berlin wants to support the project with a sum in the hundreds of millions of euros, partly to rent computing capacity in Deutsche Telekom's AI cloud, but that number is not official. The declaration also describes models that are still to be built, not systems already in use, and it does not oblige any company to share its data.

Why physical AI, and why Germany

Most of the AI boom so far has been about language: chatbots, coding assistants, search. Physical AI is the next layer. It covers models that predict how objects move, simulate machines and generate realistic training data for robots and cameras. Nvidia has been pushing this field for years with its Omniverse simulation platform and its Cosmos world models, and it needs partners with real factories and real data to make it work.

That is where Germany fits. The country's economy still rests on manufacturing, machine building, automotive and their suppliers, many of them family owned firms in the so called Mittelstand. Merz addressed those business owners directly at Kraftwerk. "The biggest advantage we have in Germany is our so-called Mittelstand," he said. "Please open your doors for AI to make use of your data." He added that he was optimistic Germany could become "one of the leading, if not the leading" countries for spreading AI across its companies.

Huang offered a reassurance that Nvidia has used in many countries. "That knowledge belongs to you," he told the audience. "It cannot be replaced. AI can only augment your capability, it will not replace your capability." According to tagesschau, he also urged Germans to treat their data as a national resource and to use it at home to build their own "national intelligence", instead of exporting their know how and later buying it back as imported AI. He called the moment a new industrial revolution and, quoting the Chancellor, said Germany was "re-engineering" itself.

Merz said he wants to meet Huang at least once a year to review progress. The message to investors at the summit, where around 200 chief executives from Germany and abroad were expected, was clear: Germany wants to be seen as open for AI business.

Siemens shows what this looks like on a factory floor

The most concrete numbers of the day did not come from the declaration but from Siemens. At its electronics factory in Erlangen, the company presented AI tools it has built together with Nvidia since the two firms announced at CES in January 2026 that they would build fully AI powered manufacturing sites. The plant makes drives and frequency converters for machine tools, has around 1,000 production staff and, according to Siemens, already runs more than 100 AI algorithms in production, logistics, quality inspection and maintenance.

The first tool tackles a slow and boring step: setting up camera based quality inspection for a new printed circuit board. Until now, engineers had to wait weeks for the first physical prototype, photograph it and manually program the inspection machines. Siemens now uses the board's digital twin and Nvidia's Cosmos world models to generate thousands of photorealistic synthetic images of good and faulty boards before the real one exists. The inspection system is trained in advance. Siemens says this reduces engineering effort by up to 85 percent and cuts time to market for new boards by up to 40 percent.

The second tool is an AI agent for production planning. When a new order arrives, it generates several production scenarios, weighs capacity, material availability, delivery dates and line setup, and shows the planner the impact of each option. Before anything changes on the line, the plan is checked against a physics based digital twin of the whole factory. Siemens says this can save up to 20 percent of costs, including capital and rush costs. A human planner still makes the final choice.

Note the phrase "up to" in all of these figures. They are best case values from a showcase plant that Siemens is turning into its global blueprint, with a 500 million euro investment in the Erlangen campus. They are not averages across German industry.

Automation as a jobs argument

Siemens board member Cedrik Neike gave the clearest economic reasoning. Speaking in Erlangen, he told Reuters that manufacturers in high wage countries need productivity gains of 7 to 9 percent a year to stay competitive. Siemens wants to use AI to raise the share of automated work in factories from 30 to 40 percent today to 70 to 80 percent. His framing is that this is how manufacturing jobs stay in Germany at all, rather than moving to cheaper countries.

That argument will be tested. Doubling automation sounds like fewer jobs per factory, and unions will want to know who benefits. But the alternative Neike describes, factories that slowly lose out to cheaper locations, is not job safe either. The honest version is that physical AI changes which jobs exist in a plant: fewer people programming inspection machines by hand, more people supervising agents and maintaining digital twins.

Nvidia's money: four German startups

Huang also brought investments. Nvidia announced planned stakes in four German startups as part of what it describes as a nearly 10 billion dollar commitment to companies, universities and research institutions across Europe. According to tagesschau, Huang said Nvidia invested almost one billion dollars in Germany last year alone. The company already backs German names such as n8n, Neura Robotics and Black Forest Labs.

The new four are a mixed group. Agile Robots in Munich builds AI powered robot arms and the Agile ONE humanoid for factories, logistics and healthcare, which fits the physical AI theme directly. Avelios Medical, also in Munich, develops an AI native hospital information system. Mirelo builds models that generate sound effects, speech and music for video. Reflex Aerospace in Berlin designs satellites that it says can be ready for launch in about 12 months instead of years. Nvidia did not disclose amounts and described all four as planned investments.

A quantum record on a German supercomputer

The same day, Nvidia researchers reported a different kind of milestone on German soil. On JUPITER, Europe's first exascale supercomputer at the Jülich Supercomputing Centre, they ran a 104 qubit simulation of a fault tolerant quantum processor, which Nvidia calls the largest and most detailed of its kind to date. The run used 8,192 GH200 chips and modelled realistic noise, including errors that build up over a circuit and qubits that leak out of their working states. It has nothing to do with factories directly, but it shows why Nvidia keeps choosing Germany as a showcase: the country has the supercomputers, the research centres and the industrial customers.

Why it matters

For Nvidia, Germany is the most important industrial market in Europe and a test case for its sovereign AI pitch: help a country build AI on its own data and its own infrastructure, with Nvidia hardware and software underneath. Deutsche Telekom's Industrial AI Cloud in Munich, opened in February 2026 with Nvidia and the data centre partner Polarise, is already part of that picture. A government declaration gives Nvidia political backing and a direct line to thousands of mid sized manufacturers that would otherwise be hard to reach.

For Germany, the deal is a signal at a moment when the country is trying to attract private investment and show that its industrial base can keep up. It also raises fair questions. Europe talks a lot about technological sovereignty, yet its flagship industrial AI initiative rests on a US company's chips and models. There was an earlier German plan for sovereign open source foundation models, called SOOFI, launched in 2025 with research institutions and AI companies, and pre summit reports suggested the Nvidia cooperation would build on it. How the two fit together has not been explained publicly.

And for the Mittelstand, the key promise is access. If small manufacturers really get Nvidia's industrial models, simulation tools and affordable compute, that could matter more than any headline. If the initiative ends up serving mainly the big groups that already work with Nvidia, such as Siemens, the declaration will have been mostly symbolic.

What we still do not know

Several important details are open. There is no official figure for public funding and no timeline for the first models. It is unclear which companies and research institutes will join the task force, how data from different firms will be protected, and on what terms small companies can use the models. It is also unclear whether the models will be open in any sense or tied to Nvidia's platform. Jensen Huang gives the keynote at GTC Berlin on Wednesday, 21 October, so more details may come there.

Dany's take

I like that this is about factories and not another chatbot. The Siemens examples are the right kind of AI: boring, measurable and useful, like training an inspection camera before the circuit board even exists. But a declaration of intent is the easy part. Germany has signed many AI strategies before, and the gap between a stage at Kraftwerk and a small machine builder in Swabia actually using a physical AI model is big. Watch three things: real money in the next federal budget, the first named Mittelstand pilots, and whether any of the resulting models are open. If those show up within a year, this was a turning point. If not, it was a good photo opportunity for both sides.

The bottom line

Germany and Nvidia have agreed to work together on AI models for robots, factories and other physical systems, and to make them available to smaller companies and researchers, not only industrial giants. Siemens showed in Erlangen that the approach already delivers real gains in a showcase plant, and Nvidia named four German startups it plans to back. What is missing so far is the detail that turns intent into impact: money, timelines and a clear answer on who gets access. The GTC Berlin keynote this week is the next place to look for it.