Nicolas Sauvage bets on the boring parts of AI

Nicolas Sauvage believes it takes four years for the best bet to become obvious – a thought he shared on stage last week at StrictlyVC’s San Francisco event, co-hosted by TDK Ventures.

It’s an idea he’s been working to prove since 2019, when he founded the business arm of the Japanese electronics giant, now holding $500 million in all four funds. AI chip startup Groq, valued at $6.9 billion during its latest funding round last fall, is a prime example of this thinking.

In 2020, before the production of the AI ​​boom made the infrastructure bet seem obvious, Sauvage wrote a check to the company, founded by Jonathan Ross – one of the engineers who created Google’s Tensor Processing Units. Groq focused on thinking from the start: the integrated heavy lifting that happens every time a model answers a question. Ross had designed his chip by building the processor first, stripping back the architecture until, as Sauvage explained, “you can’t remove one part and it still works.”

It may seem niche to some, but knowing what he did with his parent company’s constraints, Sauvage saw an asymmetry. Unlike consumer hardware, which has a natural ceiling, the need for consideration always comes with every new application and every new model. Sauvage did not know at the time that the demand for predictions would explode this year, thanks to every AI agent that organizes and makes many calls (where one question is enough).

But somehow, Ross got lucky, too. After all, the Japanese electronics conglomerate best known for magnetic tape 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 back-to-back Stanford speeches – one indicting the VC of the business, the other listing all the reasons for its failure – Sauvage, who is French and joined TDK in Silicon Valley through an acquisition, pitched the idea to the higher-ups at TDK headquarters even though he had no apparent standing to do so. (“I’m not Japanese. I don’t speak Japanese; I don’t live in Tokyo,” he told this editor.)

After refusing to take no for an answer, he finally got the green light to create a fund that was mandated to answer one question: What’s next big for TDK, and what could kill it?

The portfolio he has assembled is full of technologies that have been of widespread interest to VCs in the past year: solid-state grid inverters, sodium-ion batteries for data centers, other battery chemistries that prevent the country’s weakness of lithium and cobalt.

The discipline behind it all is the same: point to a bottle in four years, and find the inventors already working on it.

The question is what comes next. For his part, Sauvage is looking closely at physical AI — not all robots but robots that have a specific job to do. Agility Robotics, for example, in its portfolio, focuses on a single, routine task of moving items from one location to another in warehouses facing labor shortages. Another portfolio company, the Swiss ANYbotics portfolio, is building robust robots in the most dangerous areas for human workers – places where the job description must go where humans cannot. In-line is clarity of purpose. Robots Sauvage bets don’t try to do everything; instead, they do one thing honestly.

Sauvage says he has watched the computer stack change again. GPUs dominated training – massive, parallel computations for training the model. Inference chips like Groq’s are reshaping what happens when that model speaks: fast, cheap, at scale. Now, Sauvage argues, CPUs must be reinvented. They are not the most powerful or fastest chips. But they are more flexible and more suitable for branching, decision-making logic for orchestration. When an AI agent sends a task, checks its progress, and goes back through many steps, something has to manage the entire choreography. That something, increasingly, looks like a CPU.

Then there is China. A recent report from Eclipse – a business firm he closely follows – documented what Sauvage describes as “vibe generation” – rapid, AI-assisted iteration of physical hardware prototyping, to demonstrate that vibe coding in software. Chinese manufacturers, the report found, are pushing the design-build-test cycle of physical products in ways that Western supply chains are not yet equipped to match.

For Sauvage, it’s a bottleneck signal – and one he’s already continuing with TDK Ventures’ various investments. One problem that remains unsolved, he says, is intelligence. Models are developing fast enough that physical AI feels inevitable; still lacking physical smoothness to match. Countries and companies that figure out how to iterate on atoms as fast as others can iterate on code will have an advantage in productivity. That’s the wave he’s pitching TDK Ventures on today.