Huawei, China Mobile, and Baidu Co-Found NPO Optical Interconnect Standards Organization
N.R. Finch
Huawei, China Mobile Research Institute and Baidu have launched China's first Near-Package Optics (NPO) standardisation initiative with over 20 founding members; the first technical spec is targeted for Q3 2026, aiming to unify high-speed optical interconnects for AI super-nodes.
What is NPO, and why does it need a standard now?
Near-Package Optics (NPO) — placing a compact optical engine right next to the switch or compute chip so data travels as light, not electricity, over shorter distances — sits between traditional pluggable optics and co-packaged optics as a transitional technology.
This means → NPO is not the endgame, but it is the most deployable step today: lower power and higher density than pluggables, yet far easier to manufacture than full co-packaged optics.
Large-model training clusters keep pushing bandwidth, latency and power demands higher; traditional pluggable modules have already hit a power-and-density ceiling in high-density compute systems.
Who is in the alliance, and what does it cover?
The initiative is led by Huawei, China Mobile Research Institute and Baidu, under the guidance of the Global Computing Alliance.
Over 20 founding members span networking gear (ZTE, H3C), optical modules (HG Genuine, Accelink), connectors (Molex, Luxshare, Hirose), photonic chips (Zhenzheng Photonics, AINA Semiconductor), SerDes and wafer fabrication — every major link in the chain.
In plain terms = from chips to connectors to complete systems, the key players across the entire supply chain are now sitting at one table.
What will the standard actually define?
It will first set mechanical and electrical specs, environmental-reliability requirements, O&M controls and connector interfaces — core members will contribute detailed 2D and 3D connector drawings to enable cross-vendor interchangeability.
This means → once connector interfaces are open, customers are no longer locked into a single vendor's ecosystem — procurement flexibility and bargaining power both improve.
The group will also develop a linear-drive architecture — bypassing traditional DSPs to cut power and latency — targeting single-tier all-optical interconnect covering up to 1,024 accelerator cards in one super-node.
What does the timeline look like?
Q3 2026: publish the first technical specification.
Prototype interoperability and full-scenario testing of NPO and connector products will follow.
H1 2027: target is to mature the MSA framework and supplier ecosystem, then push toward scaled commercial deployment.
This means → whether China can turn this standards framework into volume-shippable products before 2027 is the critical checkpoint for domestic AI optical-interconnect progress.
What does this signal for the broader supply chain?
This reflects a shift in China's AI-compute bottleneck — from "are there enough chips?" to "how do you connect them?" — with optical interconnect emerging as the new chokepoint.
After the standard unifies, suppliers in optical modules, connectors and related segments stand to gain a larger addressable market, but also face homogenisation pressure once interfaces are open.
Put simply = whoever turns the spec into shipping product fastest gains first-mover advantage; the standard itself is a ticket to the game, not a moat.
Content is for reference only, not financial advice.