AI's Unsexy Side: Nicolas Sauvage's Bold Bet for the Future

In the whirlwind of generative AI hype, much of the focus lands on flashy applications and large language models. However, Nicolas Sauvage, the head of TDK Ventures, believes the real future of AI lies in the “unsexy” infrastructure that powers it. Sauvage posits that the most promising investments often appear obvious only in retrospect – typically four years after the initial bet. This philosophy, shared at the recent GearTech’s San Francisco event co-hosted by TDK Ventures, is driving a $500 million investment strategy across four funds, and is already yielding impressive results. He’s betting on the foundational layers, the bottlenecks, and the often-overlooked components that will truly unlock AI’s potential.

The Groq Story: A Vision Validated

A prime example of Sauvage’s foresight is his early investment in Groq, an AI chip startup founded by Jonathan Ross, a former engineer from Google’s Tensor Processing Unit (TPU) team. In 2020, long before the generative AI boom, Sauvage recognized the critical importance of inference – the process of using a trained model to respond to queries. Groq’s unique approach, building the compiler *before* the chip architecture, resulted in a highly optimized design. As Sauvage describes it, the architecture is so streamlined, “you can’t remove one part and have it still work.”

While seemingly niche at the time, Sauvage understood the inherent asymmetry between consumer hardware, which faces natural limitations, and the ever-increasing demand for inference. The explosion of AI agents requiring complex, multi-step reasoning has only validated this belief. Groq’s valuation has soared to $6.9 billion, demonstrating the power of anticipating future needs.

An Unlikely Venture Capitalist

The story of TDK Ventures itself is a testament to Sauvage’s ability to challenge conventional wisdom. A Japanese electronics conglomerate, traditionally known for magnetic tape, might not seem like a natural fit for venture capital. Sauvage, a French national who joined TDK through an acquisition, pitched the idea of a corporate venture arm despite lacking the typical credentials. “I’m not Japanese. I don’t speak Japanese; I don’t live in Tokyo,” he acknowledged.

After persistent advocacy, he secured funding for a fund with a singular mandate: to identify the next big opportunity for TDK and, crucially, what could threaten its existing business. This proactive, defensive approach sets TDK Ventures apart.

Investing in the Foundational Layers

Sauvage’s portfolio reflects this focus on foundational technologies. He’s investing in areas like:

• Solid-state grid transformers: Addressing the need for more efficient and reliable power infrastructure.
• Sodium-ion batteries for data centers: Offering a more sustainable and geopolitically stable alternative to lithium-ion.
• Alternative battery chemistries: Reducing reliance on scarce materials like lithium and cobalt.

The common thread is identifying and addressing bottlenecks – the constraints that will limit future growth. Sauvage’s strategy is to pinpoint these challenges four years in advance and invest in the companies already working on solutions.

The Rise of Physical AI and the Importance of Specificity

Looking ahead, Sauvage is particularly interested in physical AI – robotics with a clearly defined purpose. He’s not chasing general-purpose robots, but rather those designed for specific, repetitive tasks. Agility Robotics, a portfolio company, focuses on automating material handling in warehouses, addressing critical labor shortages. ANYbotics, another investment, builds ruggedized robots for hazardous environments, performing tasks too dangerous for humans.

The key, according to Sauvage, is clarity of purpose. These robots aren’t trying to do everything; they excel at one hard thing, reliably. This focused approach increases the likelihood of success and rapid deployment.

The CPU Renaissance: Orchestrating the AI Symphony

Sauvage also anticipates a shift in the compute stack. While GPUs have dominated training – the computationally intensive process of teaching AI models – inference chips like Groq’s are optimizing the response phase. Now, he believes CPUs are poised for a resurgence.

CPUs may not be the most powerful or fastest chips, but they offer unparalleled flexibility and are ideally suited for the complex, branching logic of orchestration. As AI agents delegate tasks, monitor progress, and iterate across multiple steps, a central controller is needed to manage the entire process. That controller, increasingly, will be a CPU.

China's Manufacturing Advantage: "Vibe Manufacturing"

Sauvage is closely monitoring developments in China, particularly the phenomenon of “vibe manufacturing,” as documented in a recent report from Eclipse, a venture firm he follows. This refers to the rapid, AI-assisted iteration of physical hardware prototyping, mirroring the speed of software development. Chinese manufacturers are compressing the design-build-test cycle in ways that Western supply chains are struggling to match.

This represents another critical bottleneck. Sauvage is already responding with investments designed to address this challenge. One remaining unsolved problem is dexterity – the ability to manipulate objects with precision and adaptability. He believes that the countries and companies that can iterate on physical products as quickly as others iterate on code will gain a significant manufacturing advantage.

The Iteration of Atoms

Sauvage emphasizes that models are improving rapidly, making physical AI increasingly feasible. However, the physical fluency to match this intelligence is still lacking. The ability to rapidly prototype and refine physical hardware – to iterate on “atoms” as quickly as we iterate on “bits” – will be the key differentiator. TDK Ventures is positioning itself to capitalize on this wave.

Beyond the Hype: A Long-Term Perspective

Nicolas Sauvage’s approach to AI investing is a refreshing counterpoint to the current hype cycle. He’s not chasing the latest trends, but rather focusing on the fundamental infrastructure that will underpin the future of AI. By identifying bottlenecks, investing in foundational technologies, and taking a long-term perspective, he’s building a portfolio that is poised to deliver significant returns – and shape the future of technology. His “unsexy” bet on the future of AI may well be the smartest one of all.