Sourcery: Why AI’s $10T Buildout Needs Photonics, Not Just More GPUs
AI Podcast Summaries from Transcripted.ai (VIDEO)
September 30, 2026
AI’s next bottleneck may not be software at all, but the physics of moving data fast enough.
Speakers Molly O'Shea
TopicsDaily NewsNews
Molly O'Shea (0:01)
When the AI boom collides with the limits of physics, the bottleneck is no longer just software. Today we're diving into why photonics may be essential to the next phase of computing, and why one company called TheMA is racing to meet that demand.
This is a fascinating story because it brings together two really compelling founders. Yannick De Koninck describes himself as a "nerdy kid" drawn to Lego and engineering, then spent two decades in what was once a niche field, silicon photonics. He helped build Nvidia's photonics technology to product-ready maturity. And now he's starting fresh at TheMA alongside Herwig Van Hove, who has this interesting background in investing before moving late into operations, with a long focus on deep tech and talent. So what's their central argument? It's actually quite simple. AI models have become so large that they no longer fit on a single GPU.
Multiple GPUs must be connected efficiently, and copper is running out of steam. Optical interconnects, enabled by photonics, can deliver the bandwidth modern AI demands. Right, but that shift is not just technical, it's architectural. In the age of agentic AI, there's constant communication between GPUs and tools across servers. Data centers need a flatter, denser network with lower latency. The old hierarchical tree of switches is becoming less suitable.
And then there's the energy angle. Herwig says that when CTOs are asked what keeps them awake at night, "energy consumption is the primary concern because it is existential." The speed of change, accelerated by ChatGPT's viral rise in late 2022, caught much of the industry off guard. That makes sense, but what about the supply chain? Are they ready for this shift? Not really. Foundries may handle wafer production, but finished photonic chips also require lasers, substrates, tooling, and higher throughput with better yields. Major companies like Lumentum and Coherent are scaling, but the rest of the ecosystem is still catching up.
TheMA's founding story adds a strategic twist here. Herwig and Yannick were looking for an industrial bridge between university spinouts and large-scale manufacturing when a European foundry went out of business. They acquired it and found themselves just 20 kilometers from IMEC, with access to world-class research and talent. Yannick says that leaving Nvidia felt like leaving a "golden palace" for a "pirate ship," but he saw it as a once-in-a-lifetime chance to build something new. Production is targeted for 2027, with full ramp-up by 2028
Their final message is deeply optimistic. Rather than seeing Europe as slow or committee-driven, they argue it can combine American agility with industrial depth and sovereign ambition. As Yannick puts it, "Europe is ready for this."
Thousands of transcripts fetched by people building searchable podcast archives
Fetch the whole transcript
The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000792398685