US ban on Chinese optical parts threatens AI infrastructure costs

Research firm Counterpoint warned on Wednesday that a potential US ban on Chinese optical transceiver modules would inflict collateral damage on American hyperscalers by exacerbating an existing supply bottleneck that no US firm can quickly resolve. The optical transceiver—a component that ensures speedy transmission and data integrity within and across server racks—is essential to the hundreds of billions of dollars in AI infrastructure that companies like Amazon and Microsoft are building across the country.

Chinese firms dominate this market, holding nearly two thirds of global optical transceiver supply. Key players include Zhongji Innolight Co. and Eoptolink Technology Inc., which have built a manufacturing edge in producing the high-precision modules at scale. If Chinese suppliers are cut off, US hyperscalers would face higher costs and suffer from lower utilization of expensive AI accelerators, since the infrastructure deployment would become more constrained and costly. Counterpoint analyst Neil Shah cautioned that "the belief that the optical transceiver market can be neatly divided geographically misinterprets how the hardware ecosystem operates," adding that "the global AI ecosystem remains heavily reliant on Chinese optical module vendors for scale execution."

The proposed ban would also harm Western firms in the AI supply chain. The optical transceivers made by Chinese companies integrate chips from Broadcom Inc. and Marvell Technology Inc., lasers and optical chips from Lumentum Holdings Inc., and components from Japan's Mitsubishi Electric Corp. By restricting Chinese optical modules, the US would indirectly reduce demand for these Western suppliers' components. The initial report of the trade curb triggered a stock reaction: Zhongji Innolight tanked as much as 14% on Wednesday, while Western rivals including Applied Optoelectronic Inc., Coherent Corp., and Nokia Oyj rallied—though the broader effect on them remains negative given their dependence on integration into Chinese modules.

Shah underscored the fundamental constraint: Western suppliers do not have the capability to absorb Innolight and Eoptolink's volume within the next year or two. This means that even if Western firms began ramping production immediately, they could not fill the gap in time to prevent disruption to US hyperscalers' deployment schedules. The ban thus risks creating a hardware bottleneck that slows down the buildout of AI infrastructure by the world's largest cloud operators—a costly outcome for companies and the US economy at a critical moment in AI expansion.