Sep 30, 2026 · 33 min
Photonics becomes AI infrastructure’s next manufacturing bottleneck
The $10T AI Buildout Has a Photonics Problem
As GPU clusters expand, the availability of optical interconnects and production capacity may constrain AI infrastructure more than model design does.
- 1Copper links struggle to meet the bandwidth, distance, and energy demands of increasingly large GPU clusters.
- 2The photonics supply chain lacks enough scaled manufacturing, creating an opening for a European pure-play photonic foundry.
- 3The company aims to reach production in 2027 as Europe weighs its role in AI infrastructure and advanced manufacturing.
Don't miss
The guests frame manufacturing capacity—not technical invention alone—as the decisive constraint on optical interconnects for AI factories.
The brief
Yannick André and Herwig van Hove trace how silicon photonics led them toward a company built around a basic infrastructure gap: AI clusters need faster links than copper can provide.
As models spread across more GPUs, optical links and lasers become essential for moving data across distance without compounding bandwidth and energy costs inside data centers.
The guests’ central claim is that photonics has a supply-chain problem, not merely a technology problem: too few companies can manufacture components at the required scale.
André discusses his experience at Nvidia, while both guests describe IMEC’s research role, customer demand, hiring constraints, and Europe’s opportunity to build advanced infrastructure.
The company’s target is production in 2027, making execution the test of whether Europe can turn photonics research into capacity for AI factories.
Featuring
Herwig van Hove
IMEC
Nvidia Corporation
United StatesListen to the full episode and explore every guest, topic, and moment on PodLume.