Nicolas Sauvage recently shared his insights at the StrictlyVC event in San Francisco, expressing a belief that it typically takes about four years for the most promising investments to become apparent. This perspective has been central to his work since he established the corporate venture arm of TDK, the Japanese electronics firm, in 2019. Currently, TDK Ventures oversees $500 million spread across four distinct funds.
One of the standout examples of his investment philosophy is Groq, an AI chip startup that achieved a valuation of $6.9 billion in its latest funding round. Sauvage recognized Groq's potential early on, investing in the company prior to the surge in interest surrounding generative AI. Founded by Jonathan Ross, a former engineer behind Google's Tensor Processing Units, Groq focuses on inference--the crucial computational tasks that occur when models generate responses to queries. Ross's innovative approach involved designing the chip around a compiler, ensuring that every component was essential to its functionality.
While some may have viewed Groq's focus as niche, Sauvage understood the growing demand for inference, which continues to escalate with the introduction of new applications and models. This year, the need for inference has surged, driven by AI agents that now manage multiple requests simultaneously, as opposed to the traditional single-query approach.
Interestingly, Sauvage's partnership with TDK might not have seemed like an obvious fit, given the company's historical focus on magnetic tape. However, after presenting a compelling case for corporate venture capital at Stanford, he persisted in seeking approval from TDK's headquarters in Tokyo, despite facing initial skepticism due to his background.
Since then, he has built a diverse portfolio that includes innovative technologies such as solid-state grid transformers and sodium-ion batteries, which aim to mitigate the geopolitical risks associated with lithium and cobalt supply chains. His strategy revolves around identifying future bottlenecks and supporting the founders who are tackling these challenges.
Looking ahead, Sauvage is particularly interested in the advancements in physical AI. He highlights companies like Agility Robotics, which specializes in automating tasks in warehouses, and ANYbotics, which develops robots for hazardous environments. These companies exemplify a focused approach, ensuring that their robots excel at specific tasks rather than attempting to tackle everything.
Sauvage also anticipates a shift in the computing landscape. While GPUs have dominated model training, he argues that inference chips like those from Groq are revolutionizing operational efficiency. He believes that CPUs may experience a resurgence, as they offer the flexibility necessary for managing complex decision-making processes in AI.
Furthermore, he is observing rapid developments in China, where AI-assisted hardware prototyping is significantly accelerating product development cycles. This trend poses a challenge for Western supply chains, which are not yet equipped to keep pace.
Ultimately, Sauvage sees the future of physical AI as promising, yet acknowledges the ongoing challenges in achieving the dexterity required for practical applications. The race to iterate on physical products as swiftly as software will define the next phase of manufacturing advantage.