Interview, Fireside Chat
The $10T AI Buildout Has a Photonics Problem
- Photonics is projected to become the dominant technology for GPU interconnection in AI infrastructure by 2026, driven by bandwidth limitations in copper interconnects and a global $10 trillion rebuild of data centers.
- A significant technological revolution is expected starting in 2024, with market opportunities exceeding optimistic analyst predictions and Nvidia expected to invest assets equivalent to 4x the total market size in the first six months of 2024.
- Production is scheduled to begin in 2027, with a full ramp-up targeted for 2028, following a 3-to-5 year project timeline for developing complex foundry infrastructure.
- Critical supply chain constraints involve insufficient tooling throughput, low manufacturing efficiency and yield, and a scarcity of substrates and raw materials where only a handful of suppliers currently exist.
- European industrialization is anticipated to accelerate due to decades of trained talent, available infrastructure, and the European Council's Chip Act 2 prioritizing pure play industrialization, though grid capacity shortages and permitting complexities could delay projects.
- Strategic plans include adopting an American-style decision-making culture to counter perceptions of slow European processes, aiming to build a self-sustaining ecosystem that attracts senior talent from firms like AWS, TSMC, and Lam Research.
- Market dynamics indicate a shift from academic research to industrial scale, with Lumentum and Coherent identified as the only Western entities currently scaled to meet demand, while a potential east-west supply chain divide poses risks.
- Regulatory hurdles regarding grid capacity, exemplified by Google's inability to secure power in Amsterdam after three years, necessitate a focus on energy-efficient infrastructure to enable the development of a homegrown European data center network.