Deeptech Mash: Episode 3
with David Grimm, Partner and Sebastian Hunte, Investor
In the third episode of Deeptech Mash, Dave and Seb discuss the growing importance of photonics in computing, particularly in AI and data centres. The conversation revolves around NVIDIA’s move towards optical networking and insights from Lawrence Lundy-Bryan (Lunar VC)’s recent blog on photonics. The duo delve into the benefits and challenges of optical computing, focusing on two major issues: non-linear functions and memory limitations. They also explore the concept of optical interconnects and their potential to optimise data centre performance by reducing latency and increasing bandwidth per watt. The episode concludes with a call for more investment and innovation during the paradigm shift towards an AI-first world.
Overview of the key discussed themes below, but for a detailed version watch full the conversation here:
Photonics is now core to scaling AI infrastructure
As AI models grow, data movement—not compute—is the primary bottleneck. Photonics (especially for interconnect) is emerging as the most effective solution due to its speed, bandwidth, and energy efficiency advantages over traditional electronics.
Optical interconnect is a near-term opportunity
Unlike general-purpose optical compute, which remains a long-term goal, optical interconnect is commercially viable today. Startups and hyperscalers are increasingly investing in co-packaged optics and photonic links to accelerate AI workloads and reduce energy consumption.
Energy efficiency drives infrastructure innovation
With up to 40% of AI power spent on data movement, bandwidth per watt has become a critical metric. Optimising interconnect layers directly boosts overall system performance and utilisation, making photonics a strategic priority for data centers.
Europe has a unique window to lead in deeptech
The shift toward AI-native infrastructure creates a rare chance for Europe to build global leaders in photonics, memory, and compute— if capital and local champions align.
