Optical AI Chip Companies Face Bigger Problems

Research firm B. Riley warned that Amazon and OpenAI are developing new "network flattening protocols" that could reduce demand for optical transceivers (specialized components that connect AI chips using light instead of electricity) by 40% to 50%. This is a separate concern from the NVIDIA delay but hits many of the same companies. The NVIDIA Kyber delay already meant a delayed rollout of the next generation of optical technology (called CPO). Combined with hyperscalers designing around fewer optical connections in their new architectures, companies that make optical components are facing pressure from two directions.

Names like Lumentum (LITE), Applied Optoelectronics (AAOI), and Coherent (COHR) have been valued based on continued growth in optical demand from AI data centers. Both of these developments cut against that assumption. The Amazon and OpenAI network redesigns matter because they represent hyperscalers actively engineering around optical bottlenecks rather than paying whatever it costs. When the biggest customers start designing alternatives to your product, your pricing power erodes. This is the same pattern showing up across the AI supply chain: customers who were assumed to pay any price are quietly building alternatives. Combined with the NVIDIA delay, this represents multiple pressures on the optical portion of the AI trade. It also validates the broader thesis that AI infrastructure costs are being scrutinized and optimized in ways the market has not fully priced.