Photonics: The $91 Billion Question - Layer 1 (Pt.2)

The July AI-infra correction separated disciplined portfolios from levered concentration. Our 3Q26 rebalance concludes with a clear principle: technical relevance is not enough. We exit weakened theses, favour bottleneck resolvers like Soitec, and size unproven optionality with discipline.

OCS roadmaps are splitting by workload: high-radix MEMS for scale-out and training, lower-latency silicon photonics for inference scale-up. Part 4 examines Google’s V8i push, NVIDIA’s dual CPO+OCS approach, market growth to $2.5B+, and whether microLED can solve short-reach constraints.

Part 2 of the rebalance deepens the focus on power electronics for 800V DC architectures, sharpens photonics conviction around hybrid bonding and space optics, and refines the neocloud and software allocations toward the highest-asymmetry names in the AI value chain.