Hua Hong Semiconductor Bets on Silicon Photonics, Targeting AI Optical Module Market with 2027 Capacity Expansion
nashnova research
Hua Hong Semiconductor (华虹半导体), China's second-largest foundry, will ramp silicon photonics capacity in its 2027 expansion, pivoting from legacy specialty-process foundry work toward core AI optical-module components as data-center interconnect demand surges.
What exactly is Hua Hong planning?
Chairman Bai Peng said on the November 5 earnings call that silicon photonics and power-management chips are the two priorities for "significant capacity additions" in 2027.
His words: "We want to participate in all silicon-based devices inside optical modules." This means → Hua Hong is not targeting a single component — it wants every silicon layer the module needs.
In plain terms = if an optical module uses a silicon chip anywhere inside it, Hua Hong wants that order on its production line.
Why does silicon photonics matter now?
Silicon photonics — a technology that moves data between chips using light instead of electrical signals — is the foundational layer of AI optical modules.
Data centers need ever-faster high-speed interconnects to support AI training and inference; optical modules are the "nervous system" linking servers. This means → whoever can produce more silicon-photonics devices at lower cost holds a chokepoint in AI infrastructure.
This reflects a broader trend: AI demand is pulling in not just GPUs and advanced nodes, but mature-node specialty processes too.
What does this strategic pivot signal?
Hua Hong's traditional label is "specialty-process foundry" — power devices, analog chips, embedded memory — all far from the AI supply chain.
Committing to silicon photonics means → the company is stretching from legacy foundry work into AI core-component manufacturing, trying to stake a position in the AI wave.
In plain terms = Hua Hong used to make "supporting-cast parts." Now it wants into the "lead-actor supply chain."
Can the plan deliver? Where is the risk?
Whether the 2027 expansion lands on schedule is the first question mark — at least two years separate announcement from volume production.
Whether capacity ramp-up speed matches optical-module customers' delivery cadence is the second — AI infrastructure build-out cycles are short and demand shifts fast; fall behind and customers route around you.
This reflects a challenge shared by every company chasing the AI tailwind: get the strategic direction right but miss the execution tempo, and the window closes.
市场有风险，内容仅供研究参考，不构成投资建议。