Germany's Best AI Data Is Sitting in Someone's Head

"Please open your doors for AI to make use of your data."

That was the German Chancellor this week, speaking to the Mittelstand. Not a startup founder in a hoodie. Friedrich Merz, in a former power station in Berlin, next to Jensen Huang.

A big week for AI in Germany

At the Invest in Germany Summit, the federal government signed a joint declaration of intent with Nvidia to build industrial AI in Germany and Europe. Nvidia also named investments in four German startups, including Agile Robots and Avelios Medical, as part of a wider European pledge.

Huang had a clear message for German manufacturers. "The next wave of AI is physical," he said, meaning robots, factories, cars and heavy industry. He called Germany's manufacturing data and know-how "your natural resource," and he warned: "You don't have another chance."

He also said something I want to hold on to: import the technology, but don't "export your culture, export your values, export your knowledge."

The data is not where they think it is

When politicians say "data", most people picture a big database waiting for a model. Rows and rows of clean sensor readings.

I've been in software long enough to know it rarely looks like that. The most valuable knowledge in a company usually sits somewhere much less glamorous. In a spreadsheet one person maintains. In a comment that says "do not change this value". In the head of the colleague who has been there for twenty years and knows why the machine on line three gets a different setting in winter.

That is the real natural resource. And you can't feed it to any model, because nobody ever wrote it down.

AI can only use what you know if you know it out loud.

This is not only a factory problem. Every engineering team I know has the same thing. The service only one person dares to deploy. The incident that happened twice and was fixed twice, by the same person, from memory. We joke about the "bus factor" and then do nothing about it.

For years that was a slow risk. Now it's a fast one. A team that has written down its decisions, its odd edge cases and the reasons behind them can hand that context to an agent tomorrow. A team that hasn't will get confident, generic answers that sound right and miss the one detail that mattered.

What happens next

I expect a lot of money and a lot of announcements. Some of it will work, some of it will be slideware. The companies that win won't be the ones with the biggest GPU contract. They will be the ones who did the boring work first: writing down what their experts know while those experts are still around to ask.

And Huang is right about one thing. If that knowledge gets written down, it should stay ours. It belongs in the company's own documents, not scattered across prompts in someone else's tool.

Where I would start

If you are an engineer:

1. Pick one thing only you know how to do, and write it down this week. A short note is enough: what, why, and the trap.
2. When you fix a strange bug, add one sentence explaining why the fix works. Future you, and any agent, will thank you.
3. Ask the oldest person on your team one question a week about why something is the way it is. Write the answer next to the code.

If you lead a team:

1. Treat documentation as real work. Put it in the sprint, review it, and celebrate it.
2. Pair your most experienced people with newer ones on purpose. Knowledge moves best between people first, and onto paper second.
3. Before buying any AI tool, ask a simple question: what would we actually give it to read?

Germany has decades of hard-won expertise. That is not nothing, and plenty of countries would love to have it. But a natural resource only matters once you dig it up. For most of us, digging starts with a pen, not a GPU.

Sources: AI News: Germany and NVIDIA sign industrial AI declaration in Berlin, AP via BNN Bloomberg: Nvidia CEO urges German manufacturers to adopt AI, CNBC: German chancellor urges businesses to embrace AI