TDK Ventures’ Nicolas Sauvage on betting on the boring parts of AI

Speaking last week at StrictlyVC’s San Francisco event, Nicolas Sauvage, managing director of TDK Ventures, shared his belief that it takes four years for the best bets to look smart. Sauvage has been testing this theory since 2019, when he founded the corporate venture arm of the Japanese electronics giant TDK. Today, the firm manages $500 million across four funds. The origin of the fund was highly unlikely; Sauvage, who joined TDK in Silicon Valley, pitched the idea directly to headquarters in Tokyo despite having no traditional standing. “I’m not Japanese. I don’t speak Japanese; I don’t live in Tokyo,” Sauvage said, reflecting on his pitch to establish a venture arm tasked with finding the next big thing for the company.

The highest-profile example of this strategy is the AI chip startup Groq, which was valued at $6.9 billion during its most recent funding round last fall. Sauvage wrote a check into the company in 2020, well before the generative AI boom. Founded by Jonathan Ross, Groq focused from its inception on inference—the computational process when an AI model responds to a query. Ross designed the chip by building the compiler first, stripping the architecture down until, as Sauvage describes it, it is impossible to remove a single part and have it still function. While the bet seemed niche at the time, Sauvage recognized that demand for inference would compound as AI applications scaled.

Today, Sauvage is applying this bottleneck-focused discipline to physical AI—robots with specific, task-oriented jobs. The TDK Ventures portfolio includes Agility Robotics, which focuses on moving goods in warehouses, and Swiss portfolio company ANYbotics, which builds ruggedized robots for hazardous environments. Alongside physical AI, Sauvage is tracking shifts in the compute stack, the hierarchy of hardware used for AI. While GPUs have dominated model training, Sauvage argues that CPUs are due for a renaissance. He believes CPUs are uniquely suited to manage the complex, branching orchestration logic required when AI agents coordinate multi-step tasks.

Sauvage is also monitoring how hardware is prototyped. He points to a recent report from the venture firm Eclipse detailing vibe manufacturing, the rapid, AI-assisted iteration of physical hardware prototyping. According to the report, manufacturers in China are compressing design-build-test cycles in ways that Western supply chains cannot currently match. For Sauvage, this gap represents a critical bottleneck. He believes the countries and companies that learn to iterate on physical hardware as quickly as others iterate on software will secure a significant manufacturing advantage, a wave he is positioning TDK Ventures to ride.

Why it matters

Sauvage’s investment strategy focuses on identifying long-term bottlenecks in infrastructure and physical AI, positioning TDK Ventures to capitalize on shifts in the compute stack and manufacturing efficiency.