The AI boom is making older infrastructure bets look newly timely
Nicolas Sauvage is making a case for a part of the AI market that rarely gets the same attention as consumer-facing products: the hardware, compilers, and energy systems that make large-scale computing possible. Speaking at StrictlyVC’s San Francisco event, the TDK Ventures founder argued that strong venture bets often need years before the market recognizes them as obvious. His own benchmark is four years, and he has been trying to prove that idea since 2019, when he launched the corporate venture arm of Japanese electronics company TDK.
That patience now looks better timed than it did at the start. TDK Ventures manages $500 million across four funds, and one of its most visible wins is Groq, the AI chip startup that reached a $6.9 billion valuation in its most recent funding round last fall. Sauvage backed the company in 2020, well before generative AI pushed inference infrastructure into mainstream investor conversation.
Why inference mattered before it became fashionable
Groq was focused from the beginning on inference, the computing work performed each time a model answers a prompt. That emphasis is important because the economics of AI are increasingly shaped not only by training frontier models, but by serving them repeatedly at scale. As more products embed AI into search, coding, customer service, and automation, the number of inference calls grows with them.
Sauvage saw that demand curve early. According to TechCrunch’s account of his remarks, he understood that inference would keep compounding as new applications and models multiplied. The current rise of AI agents has only sharpened that logic. A task that once required a single model response may now involve a system planning, executing, and checking work across many calls. That shift creates pressure for faster, more efficient inference infrastructure.
Groq’s technical approach also fit the thesis. Founder Jonathan Ross, previously one of the engineers behind Google’s Tensor Processing Units, built the compiler first and then reduced the chip architecture until no part could be removed without breaking the system. To a generalist investor, that could have looked narrow. To Sauvage, it looked asymmetric.
A corporate venture arm built against the odds
There is also an unusual institutional story behind the fund itself. TDK is best known for electronics and magnetic tape, not as an obvious Silicon Valley venture sponsor. Sauvage himself described the creation of TDK Ventures as unlikely. He joined TDK in Silicon Valley through an acquisition, then pushed the company to create a venture arm despite lacking the conventional profile of a Tokyo insider. By his own account, he is French, does not speak Japanese, and does not live in Tokyo.
He persisted after internal resistance and eventually won approval to build a fund around a simple strategic question: what is the next big thing for TDK, and what could threaten it? That framing helps explain why the portfolio extends beyond pure-play AI startups. The point is not only to chase growth, but to identify enabling technologies and disruption risks early enough to matter.
The portfolio points to a wider industrial reset
The companies and technologies Sauvage highlighted suggest that the AI investment cycle is broadening into power systems, materials, and industrial infrastructure. His portfolio includes solid-state grid transformers, sodium-ion batteries for data centers, and alternative battery chemistries designed to reduce exposure to lithium and cobalt supply chains.
Those areas share a common trait: they address bottlenecks that become more visible as computing demand rises. If AI inference grows the way many investors expect, then data-center power, grid integration, and battery supply resilience move closer to the center of the technology stack. In that light, the “boring parts” of AI are not peripheral. They are part of the system that determines whether AI can scale economically and geopolitically.
That does not mean every early infrastructure bet will work. These markets are capital intensive, technically demanding, and slower to validate than software. But the logic behind the strategy is increasingly hard to dismiss. The recent AI wave has rewarded investors who spotted not just model builders, but the less glamorous technologies beneath them.
Why this matters now
Sauvage’s argument lands at a moment when AI investing is looking for its next layer. The first phase of excitement centered on models, chatbots, and application startups. The next phase is exposing the limits of power delivery, supply chains, and hardware efficiency. That is where his thesis becomes more relevant: the most valuable opportunities may still sit in markets that look unexciting until a larger shift makes them unavoidable.
For TDK Ventures, that means betting on infrastructure years before consensus catches up. For the broader market, it is a reminder that technology cycles are often won in places consumers never see. AI may be the headline, but inference chips, grid hardware, and battery chemistry are increasingly part of the real story.
This article is based on reporting by TechCrunch. Read the original article.
Originally published on techcrunch.com
