Huawei Pushes NPO Optical Interconnect Standard, Challenging Nvidia and Broadcom's CPO Roadmap
nashnova research
Huawei debuted its 7.2T near-package optics (NPO) solution at CIOE and is pushing it as an industry standard, directly challenging the co-packaged optics (CPO) path already in mass production at Nvidia and Broadcom — a standards battle that will shape the next round of AI infrastructure supply chains.
What are NPO and CPO actually fighting over?
The AI bottleneck is shifting from "how fast one chip runs" to "how fast chips talk to each other." Optical interconnect — using light instead of electrical signals to move data between chips — is the consensus direction.
CPO (co-packaged optics) puts optical components directly inside the chip package, cutting transmission latency and signal loss. Nvidia's and Broadcom's CPO solutions entered mass production in H2 2025; TSMC and Intel are also developing their own.
NPO (near-package optics) takes a similar light-meets-silicon approach but places optical components on the printed circuit board, not inside the chip package. This means → the core disagreement is not "whether to use light" but "how close to the chip the optics should sit."
What gives Huawei's NPO a fighting chance?
NPO's first selling point is a lower manufacturing bar: components sit at the PCB level, reusing mature, existing supply chains without chip-package-grade precision processes.
Its second is hot-swappability — a failed optical component can be replaced on its own. With CPO, a single component failure can take down the entire chip. In plain terms = NPO lets you swap one part; CPO risks scrapping the whole module.
Huawei is lobbying through the Optical Internetworking Forum (OIF) to make NPO an industry standard, engaging cloud vendors and component suppliers from the U.S., Japan, China, and Canada.
How does the industry read these two paths?
Tiger Ninomiya, Amphenol's director of optical standards and an OIF member, told Nikkei Asia: "We see NPO as an evolutionary step toward CPO — meeting industry needs with currently more feasible technology while addressing manufacturability, ecosystem risk, and interoperability."
This means → at least some industry players position NPO as a transitional solution, not a CPO replacement — use NPO to solve today's problems, then potentially converge on CPO long-term.
Yet NPO's standard has not yet passed OIF review, while CPO's global standard is already established through an OIF implementation agreement. This reflects Huawei's catch-up position in standards-setting influence.
Who stands to win in an NPO ecosystem?
Chinese optical-module makers Innolight (新易盛) and Eoptolink (亿芯) already play significant roles in global optical-communications and AI data-center supply chains, making them potential NPO ecosystem participants.
Yole forecasts the optical-module market will reach $112 billion by 2031, growing at a 30% CAGR.
This means → regardless of which path wins, growth certainty for the optical-interconnect sector is high. But whether a company bets on the NPO or CPO supply chain will determine its share of that $112 billion.
Can NPO genuinely challenge CPO's dominance?
BULL
Lower manufacturing bar
Reuses existing PCB processes — far easier to mass-produce than CPO.
Hot-swappable, lower maintenance cost
Individual components can be replaced without scrapping the module.
OIF channel already open
Huawei is actively lobbying within OIF; multi-country vendors engaged.
BEAR
CPO standard already locked in
CPO's global standard passed OIF implementation — first-mover advantage is real.
Nvidia and Broadcom already shipping
CPO is in mass production as of H2 2025; NPO is still at the demo stage.
May only be a stepping stone
An OIF insider explicitly called NPO 'an evolutionary step toward CPO.'
In plain terms = NPO's edge is 'available now, easy to make, easy to fix'; CPO's edge is 'more advanced, already shipping, standard already set.' The outcome hinges on which way the supply chain leans and how fast the standards review moves.
Why does this standards battle matter?
The center of gravity in AI infrastructure is shifting from single-chip performance to rack- and cluster-level communication capacity. In plain terms = the race used to be "how powerful is one chip"; now it is "how fast can a swarm of chips talk to each other."
At its core, the optical-interconnect standards war is about who defines the communication architecture underneath next-generation AI data centers. This reflects a broader shift: competition in the chip industry has expanded from "compute power" to "connectivity."
By entering the standards fight with NPO, Huawei — even if NPO turns out to be only a transitional solution — is buying China's supply chain a window to help write the rules of the optical-interconnect race.
市场有风险，内容仅供研究参考，不构成投资建议。