The four companies were named on stage in Berlin on October 20 at Nvidia’s GTC conference: Agile Robots, Avelios Medical, Mirelo and Reflex Aerospace. All are German. All are early-stage. The checks are part of a European investment program the company has described as approaching 10 billion dollars, and Nvidia did not disclose what any of the four will receive.
That omission is the most informative part of the announcement. Nvidia has spent two years converting its balance sheet into an instrument of demand, taking small positions in companies that will then buy its accelerators, and the size of a cheque is usually less important than the access that comes with it. A startup that takes Nvidia’s money also gets early allocation of hardware, engineering help and a logo it can put on a slide.
The four fit different parts of that pattern. Agile Robots, based in Munich, builds industrial robotic systems, including arms sensitive enough to handle assembly work that has to be done by feel. Avelios Medical, also in Munich, writes clinical software for hospitals and was founded by researchers from the Technical University of Munich.
Mirelo, in Berlin, works on generative models for video and sound. Reflex Aerospace, also in Berlin, manufactures small satellites on short production cycles.
None of them is a model developer in the sense that Anthropic or OpenAI is. They are users of models and, in three of the four cases, builders of physical systems that need inference at the edge: a robot arm that has to identify a part, a satellite that has to process an image before downlink, a hospital system that has to triage a scan. That is the market Nvidia sells into when it talks about industrial artificial intelligence.
The German government’s part of the day was a separate agreement on industrial AI, reported by Handelsblatt as involving more than 100 million euros of federal money, with Nvidia contributing engineers and chips rather than cash. The ratio is worth stating plainly: roughly 100 million euros of public funds alongside a corporate program described as close to 10 billion dollars.
What the public money buys is not compute. It buys a framework in which German industrial companies, research institutes and the government can coordinate on standards, data formats and pilot projects, with Nvidia inside the tent. For a government that has watched its machine-tool and automotive suppliers lose ground in software, the appeal is access to the company that currently sets the pace of the hardware.
For Nvidia, the appeal is different and older. A chip company with a dominant position has two risks: that demand concentrates in a handful of American cloud providers, and that governments decide that depending on a single foreign supplier is a strategic problem. Investing in European startups, funding European research and appearing at a German government summit addresses both. It diversifies the customer base and it makes the company look like part of the local industrial base rather than a vendor above it.
The European program has been taking shape for a year. Nvidia has backed Mistral AI in France, worked with Nebius on capacity in Finland, and signed agreements with several governments on sovereign AI infrastructure. The German announcement follows that template: a headline number, a list of partners, a public commitment from a minister, and no disclosure of what any individual company received.
The list of four is small against the money described. If the nearly 10 billion dollar figure is accurate, four undisclosed early-stage positions would be a rounding error within it, which suggests the capital is going somewhere else: data center partnerships, a cloud region, joint ventures with industrial groups, or commitments tied to hardware purchases rather than equity.
That is the pattern to watch. Nvidia’s largest European commitments have been to capacity, not to startups, because capacity is what converts into chip orders. A German AI cloud built on Nvidia hardware, or a partnership with a German industrial group to supply factory-scale computing, would move more revenue than a dozen seed rounds.
Berlin’s interest is in the politics of dependence. The German government has spent two years arguing that Europe needs its own computing capacity, and it has committed public money to build it, with limited results. The country’s largest AI companies train their models on American clouds. A summit at which an American chip company announces investments in German startups is, from the government’s side, a way of being visible in a market it does not control.
There is a version of this that compounds. A robot company with Nvidia engineering support and preferential hardware access can ship a product two years earlier than a competitor without either. A satellite manufacturer that can run inference on orbit sells a different product than one that downlinks raw imagery. Those advantages are real and they are difficult for a competitor to match when the supplier is also an investor.
There is also a version in which the announcement is mostly a photograph. Four undisclosed cheques, a government agreement worth 100 million euros, and a 10 billion dollar program that has been described in similar terms in three countries. The test is not the press release. It is whether any of these companies is still independent and larger in three years, and whether the computing they use sits on German soil or in a data center in Virginia.
Nvidia’s own interest in that last question is limited. It sells the same chips wherever they are installed, and a German cloud built with public money is as good a customer as an American one. What Berlin wants is a domestic supply chain. What Nvidia is offering is a position in its supply chain, which is a different thing, and the difference will become clearer as the money is actually spent.
