AI chips – Nicolas Sauvage at TDK Ventures argues that AI’s biggest opportunities lie in today’s “boring” infrastructure, especially inference compute.
AI’s next winners may not look exciting at first glance, and Nicolas Sauvage is building his investing strategy around that idea.
In MISRYOUM’s business lens. Sauvage’s thesis is that the most profitable bets in emerging technology tend to take years before they feel “obvious.” He shared this approach at a StrictlyVC event in San Francisco hosted by TDK Ventures. describing how corporate venture investing often requires patience to spot which capabilities will matter once demand accelerates.
MISRYOUM insight: In fast-moving AI cycles, infrastructure investments can appear niche early on, yet they often become the foundation for later breakthroughs across products, platforms, and business models.
Sauvage has been testing that theory since 2019. when he launched TDK Ventures. the corporate venture arm of Japan’s TDK. and began managing capital across multiple funds.. His most high-profile example is Groq. an AI chip startup that initially targeted inference. the compute required when models respond to real-world queries.. Unlike training. inference is tied to each interaction. meaning demand can intensify as more applications come online and as queries become more complex.
MISRYOUM’s view: The key shift here is timing. Inference-first infrastructure can look like a side bet during early AI adoption, but it may become the cost and performance driver once usage patterns expand from single prompts to multi-step tasks.
What stands out in Sauvage’s story is the method behind the portfolio.. He describes a discipline of looking for bottlenecks years ahead, then backing founders already working on them.. That process has led TDK Ventures toward areas that are only recently gaining broader investor attention. including energy storage technologies and other system-level innovations tied to industrial constraints.
In physical AI, Sauvage is also looking beyond general-purpose robotics.. His focus is on robots built for specific. high-clarity jobs. such as warehouse movement or operating in hazardous environments where human labor is limited.. The common thread, as he frames it, is reliability over breadth: doing one hard task well and repeatedly.
MISRYOUM insight: Specialization can be a competitive advantage in robotics and industrial AI, because narrow performance targets are easier to validate, scale, and refine than “do everything” visions.
Finally, Sauvage points to two more watch areas.. First is the compute stack. where he sees roles shifting from GPU dominance in training toward broader needs in orchestrating agent-like behavior. where different chip types may matter.. Second is manufacturing speed: he highlights the idea that some regions may be compressing design-build-test cycles for hardware in ways that could translate into a long-term production advantage.
For MISRYOUM, the takeaway is straightforward: Sauvage’s strategy is less about chasing hype and more about identifying what will quietly become essential once AI moves from experiments to everyday deployment.