Nicolas Sauvage is betting on the boring parts of artificial intelligence

Nicolas Sauvage believes it takes four years for the best bets to become clear – believing as he took to the stage last week at a StrictlyVC event in San Francisco, which TDK Projects Co-hosted.

It’s a theory he’s been working to prove since 2019, when he founded the investment arm of the Japanese electronics giant, which now manages $500 million across four funds. Start the AI ​​chip Your puppyworth $6.9 billion During its latest funding round last fall, it was the most striking example of this thinking.

In 2020, before the generative AI boom made the infrastructure stakes seem clear, Sauvage wrote a check to the company, which was founded by Jonathan Ross — one of the engineers who built Google’s Tensor processing units. From the beginning, Grock focused on inference: the heavy computational work that happens every time a model responds to a query. Ross designed his chip by first building the compiler, then abstracting the architecture until, as Sauvage describes it, “You can’t remove a single part and keep it working.”

It may have seemed convenient to some, but when Sauvage learned what he had done about the constraints of his parent company, he saw the asymmetry. Unlike consumer devices, which have a natural ceiling, the demand for inference continues to increase with each new application and each new model. Little did Sauvage know then that the demand for inference would increase this year, thanks to each AI agent planning and acting across dozens of calls (where one query is enough).

But in some ways, Ross was also betting. After all, the Japanese electronics group famous for its magnetic tapes is not, on the face of it, the most obvious investment partner. In fact, Sauvage describes the existence of TDK Ventures as highly unlikely. But after two successive lectures at Stanford — one explaining the case for corporate venture capital, the other cataloging all the reasons it fails — Sauvage, a Frenchman who joined TDK in Silicon Valley through an acquisition, pitched the idea to the Tokyo headquarters despite having no clear standing to do so. (“I’m not Japanese. I don’t speak Japanese, and I don’t live in Tokyo.”)

After refusing to take no for an answer, he finally got the green light in 2019 to create a fund whose mission is to answer one question: What’s the next big thing for TDK, and what might kill it?

The portfolio he has since assembled is filled with technologies that have become more broadly interesting to venture capital firms over the past year: solid-state grid inverters, sodium-ion batteries for data centers, and alternative battery chemistries that sidestep the geopolitical fragility of lithium and cobalt.

The discipline behind it all is the same: identify the bottleneck after four years, then find founders who are already working on it.

The question, of course, is what’s next. For his part, Sauvage keeps a close eye on physical AI, not all robots, but robots that have a very specific job to do. Agility RoboticsFor example, he focuses his portfolio on the single-mundane task of moving things from one place to another in warehouses facing workforce shortages. Another wallet company, Swiss Wallet Any boutiquesbuilding powerful robots to work in environments too dangerous for human workers — places where the definition of a job is to go where people can’t. The dividing line is clarity of purpose. The robots Sauvage is betting on don’t try to do everything; Instead, they do one difficult thing reliably.

Sauvage says he’s also watching the compute stack shift again. GPUs have dominated training, which is the massive, parallel computation for teaching a model. Inference chips, like Groq’s, reshape what happens when that model speaks: faster, cheaper, and at scale. Now, says Sauvage, CPUs are destined for a renaissance. It’s not the most powerful or fastest chip. But it is the most flexible and best suited to branching coordination and decision-making logic. When an AI agent delegates a task, checks its progress, and iterates through dozens of steps, something has to manage the entire choreography. This thing, increasingly, looks like a CPU.

Then there is China. A recent report from Eclipse — a venture firm he follows closely — documented what Sauvage describes as “dynamic manufacturing” — the rapid, AI-assisted iteration of physical hardware prototypes, mirroring what dynamic programming did for software. The report found that Chinese manufacturers are compressing the design, build and testing cycle of physical products in ways that Western supply chains are not yet equipped to match.

For Sauvage, this is a bottleneck signal — one he’s already moving on with various TDK Ventures investments. One remaining unsolved problem, he says, is ingenuity. Models are improving fast enough that physical AI seems inevitable; What is still missing is proper physical fluency. Countries and companies that figure out how to iterate on atoms as quickly as others iterate on code will enjoy an industrialization advantage.