While most venture capitalists flock to chatbots and generative AI apps, TDK Ventures President Nicolas Sauvage has spent five years quietly building a portfolio around the humble backbone of AI. His bets on AI chip maker Groq, robotics companies Agility Robotics and ANYbotics, and other infrastructure strategies suddenly look prescient as the industry wakes up to the hard truth. The fancy AI models everyone is chasing require serious hardware, power, and physical automation to really scale.
There’s a new obsession in the venture capital world, and it’s decidedly unglamorous. Nicolas Sauvage, president of TDK Ventures, has been preaching the gospel of AI infrastructure since 2019. At the time, most investors still didn’t understand what the Transformer model was. Now, with data centers and chip supplies tight due to the AI boom, his portfolio looks like a roadmap everyone else is rushing to follow.
Sauvage’s argument is simple but contrarian. Forget about the application layer. The real money and influence is in the technology that physically enables AI. That means chips that can actually process inference workloads, robots that can operate in real-world situations, and power systems to keep it all running. This is the kind of deep technology that makes venture capitalists nervous. Long development cycles, capital-intensive scaling, technology that doesn’t work well at conferences, etc.
But the portfolio he built tells a different story. Groq, an AI chip startup founded by former Google engineers, represents a bet on infrastructure that’s really paying off. While Nvidia dominates training workloads, Groq’s language processing unit targets inference (the actual deployment of AI models) with a completely different architecture. The company’s deterministic approach promises predictable delays. This is a critical capability for real-time AI applications that current-generation GPUs have difficulty delivering consistently.
The investment in robotics reveals another layer of Sauvage’s strategy. Agility Robotics, makers of the humanoid robot Digit, aren’t just making cute demos. The company’s machines are already operating in warehouses, processing the types of repetitive tasks that require a physical form for AI to perform. ANYbotics, on the other hand, focuses on quadrupedal robots for industrial inspection. This is another unglamorous but huge market where AI meets the physical world.
What looked like a distributed bet in 2019 now resembles a coherent infrastructure stack. As generative AI moves from research labs to production deployments, companies run into the same wall. Chip shortages, power constraints, and the challenge of translating AI insights into physical behavior. TDK Ventures positioned itself at each chokepoint before the traffic arrived.
Timing is important. In 2023 and 2024, the broader VC community pursued companies building large language model applications and AI wrappers, i.e., thin layers on top of OpenAI or Anthropic APIs. But as these markets become commoditized and margins compress, investors are rediscovering the infrastructure layer. The technology Sauvage backed in its early days has suddenly become a hot category, with late-stage venture capitalists and corporate investors weeding out companies they had previously dismissed as too technical or capital-intensive.
TDK’s corporate backing gives Sauvage an advantage over traditional VCs. As the venture arm of TDK Corporation, a Japanese electronics conglomerate with deep expertise in materials science and components, the fund is able to provide portfolio companies with more than just capital. This strategic alignment is especially important in hardware and robotics, where go-to-market strategies often rely on manufacturing partnerships and supply chain access.
The common denominator in this portfolio is not AI itself, but the enabling technologies that will determine whether AI remains an object of research or transforms industry. A power management system that prevents data center meltdowns. Thermal solution for high-density chip packaging. Sensors that allow robots to navigate unpredictable environments. These are the boring parts. The infrastructure layer doesn’t generate the breathtaking headlines, but it determines which AI applications actually scale.
Other investors are also paying attention. The past year has seen a proliferation of infrastructure-focused funds, from established companies starting dedicated hardware operations to new funds targeting AI stacks below the model layer. Chipmakers, cloud providers, and the corporate venture arms of conglomerates are all cycling through the same category that TDK entered a few years ago. The competition tests Mr. Sauvage’s theory, but it also drives up valuations and intensifies competition for deal flow.
When it comes to infrastructure investments, technological reliability is often what separates first movers from latecomers. Evaluating new chip architectures or robot control systems requires domain expertise that most generalist VCs don’t have. Because of TDK’s electronics and materials heritage, Sauvage’s team is well-versed in technology that would seem like a black box to traditional software investors. Technical depth is important when technology risk dominates business model uncertainty before a startup can generate revenue.
This strategy comes with risks that traditional VCs are rightly aware of. Hardware expansion is different from software, in that each additional customer requires physical manufacturing as well as server capacity. Robotics faces the challenge of operating in environments that are not compliant with regulatory hurdles, safety certifications, and training data. The AI chip competes with Nvidia’s large installed base and developer ecosystem. These are not risks that Sauvage ignores. These are the reasons why competition remains limited long enough for TDK to establish itself.
As AI infrastructure bottlenecks worsen, the tedium that Sauvage targeted is becoming a hot topic. Data center operators are scrambling to acquire power solutions. Cloud providers are designing custom chips to reduce their dependence on Nvidia. Manufacturers are finally piloting robots that work reliably enough in production environments. The technology that venture capitalists overlooked while chasing chatbots and image generators turned out to be the necessary foundation for everything else.
Venture capital strategies around AI are being rewritten in real time, and the unglamorous infrastructure layer suddenly becomes the main plot. Sauvage has been betting on chips, robotics and power systems for five years, and TDK Ventures is at the center of the constraints that determine which AI applications actually matter. As the industry moves from proof-of-concept demos to large-scale production deployments, the boring parts become less boring. They became the only part that mattered, and investors who realized that early on are now sitting in portfolios that everyone else wants to emulate.
