Sourcery
Sourcery

The $10T AI Buildout Has a Photonics Problem

September 30, 2026

AI Summary

5 min read

The $10T AI Buildout Has a Photonics Problem

When ChatGPT went viral in November 2022, it triggered a realization across the data center industry: the old way of moving data between chips was about to hit a wall. Copper interconnects, which had served computing for decades, were running out of steam. The models had grown too large to fit on a single GPU, and the traffic patterns inside data centers had fundamentally changed. What was needed was a shift to photonics—using light instead of electricity to move data. But the supply chain was not ready.

Why copper is dying and photonics is the only path forward

The core problem is simple arithmetic. AI models no longer fit on a single GPU, so multiple GPUs must be interconnected to run them. As GPUs get faster, they need more data at faster rates, and copper cannot keep up. But the shift runs deeper than just replacing one cable with another. Traditional hyperscale data centers were built for a different kind of traffic: a user sends a query, it hits one computer, and the answer comes back. Modern AI factories look nothing like that. With agentic AI, a model on one server calls a tool on a different server, which returns results to the GPU. There is massive traffic within the data center, all of it sensitive to latency—the time between when calculations can run. This requires a much flatter network, with far more direct conn

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

  • 1 (00:41) **Introducing the Guests and the Photonics Thesis** - Yannick and Herwig join to discuss their new venture, Pima, and the urgent need for photonics in the AI buildout.
  • 2 (01:05) **Background: Why Silicon Photonics?** - Yannick explains his 20-year journey into what was once a niche field with no job prospects.
  • 3 (03:26) **The Core Problem: Why Copper is Dying and Photonics is the Killer App** - The fundamental shift from copper interconnects to optics in AI data centers is explained.
  • 4 (05:12) **The $10 Trillion Rebuild: Data Centers vs. AI Factories** - The architectural difference between old hyperscale data centers and modern AI factories is detailed.
  • 5 (07:54) **The Speed of Change and the Supply Chain Shock** - ChatGPT’s viral launch created a demand spike that the photonics supply chain was completely unprepared for.
  • 6 (09:48) **The State of the Photonics Market** - A look at the current players and the structural gaps in the industry.
  • 7 (12:07) **The European Sweet Spot and the Founding of Pima** - Why Europe is uniquely positioned to solve the supply chain crisis.

+ Full timestamped outline available in the app

Show Notes

AI's next infrastructure bottleneck? It's everything connecting them.

As AI models grow, they increasingly need to run across massive clusters of GPUs. Traditional copper interconnects are running out of steam, pushing the industry toward optical interconnects and silicon photonics.

I sat down with Thema's Herwig Van Hove and Yannick De Koninck in Monaco to go deep on the photonics supply chain powering the next generation of AI infrastructure.

Yannick spent five years at NVIDIA, where he helped build its silicon photonics technology from scratch to product-ready maturity. He left what he calls the “golden palace” to join Thema and help build large-scale photonics manufacturing in Europe.

Herwig comes from deep tech investing and saw a different problem. The innovation was ready, but the industrial supply chain wasn't prepared for the speed or scale of AI demand.

We get into

› Why copper is running out of steam inside AI infrastructure

› How photonics could reduce the networking and energy bottlenecks facing AI factories

› Why the photonics supply chain wasn't prepared for the AI boom

› What Yannick learned building silicon photonics at NVIDIA

› Thema's plan to begin production in 2027 and build a major photonics manufacturing platform in Europe

Thema believes the opportunity extends far beyond a single component.

“If you look at an AI factory today, half of that is GPUs, but the other half is the technology to interconnect these GPUs.”

Herwig believes Europe has a rare opportunity to turn decades of photonics research into industrial scale, pointing to ASML as proof that Europe has done it before.

Recorded at the Strike x Sourcery Summit in the South of France


Herwig Van Hove: https://www.linkedin.com/in/herwigvanhove

Yannick De Koninck: https://www.linkedin.com/in/yannick-de-koninck-63304314

Molly O’Shea: https://x.com/MollySOShea 

Sourcery: ⁠https://x.com/sourceryy


𝐄𝐏𝐈𝐒𝐎𝐃𝐄 𝐋𝐈𝐍𝐊

YouTube: https://youtu.be/HfSIbxNv-To


𝐒𝐏𝐎𝐍𝐒𝐎𝐑𝐒

• Brex—The modern finance platform, combining the world’s smartest corporate card with integrated expense management, banking, bill pay, & travel. Sourcery

More from this podcast

 Sourcery →