China's optical modules and domestic AI chips: building the sovereign compute stack from the bottom up

China's optical module manufacturers and PCB producers are reporting strong demand as domestic AI chip clusters scale up. It's the unglamorous infrastructure layer that makes Huawei Ascend clusters viable without Nvidia—and the supply chain is materializing.

In plain terms: To build AI clusters without Nvidia chips, China needs its own version of every component in the stack - chips, memory, interconnects, networking, power. The optical module and PCB layer is the part that's actually shipping now.

The fact

China's optical module manufacturers and PCB producers are reporting strong demand driven by domestic AI chip cluster deployments, according to Pandaily. Optical interconnects - the high-speed fiber links that connect chips within and between servers - are a critical layer in large-scale AI training clusters, where data movement between GPUs is a primary performance bottleneck.

Our take

The Nvidia export control story focuses on chips. The less-covered story is what happens below that layer: interconnects, switching fabric, power delivery, and PCBs. A Huawei Ascend 910C cluster can only scale to match H100 performance if the optical interconnect and switching layers perform comparably. Chinese optical module manufacturers - Innolight, Eoptolink, and others - have been quietly building capacity in this tier for two years. Strong demand from domestic AI deployments means those clusters are actually being assembled and connected, not just announced. The sovereign compute stack is being built from the bottom up, and the optical layer is now a production reality rather than a dependency risk.

What to watch

Whether Chinese optical module vendors start winning contracts from non-Chinese hyperscalers - which would signal that the sovereign compute build-out has produced globally competitive component suppliers, not just captive demand.