Huawei pitches near-packaged optics as co-packaged costs bite

Huawei demonstrated a 7.2Tbps near-packaged optics (NPO) module for its next-generation SuperPOD AI infrastructure, claiming a reduction in optical-interconnect latency to 10ns from 100ns and roughly 60% lower power consumption versus conventional DSP-based pluggable modules. Huawei positions NPO as a more manufacturable and serviceable alternative to co-packaged optics, while an Optical Internetworking Forum effort is developing a competing 6.4Tbps interface that could scale to 12.8Tbps. Adoption remains uncertain amid competing CPO and Arista XPO approaches, interoperability questions, and Huawei's exposure to US export restrictions.

Analysis

The investable implication is not a near-term revenue event but a potential architecture fork in AI networking. NPO shifts DSP functionality toward the switch ASIC and away from the detachable optic, which could pressure the long-duration content-per-port thesis for merchant optical DSP suppliers if adopted at scale. Conversely, switch vendors able to control ASIC, board design, thermal engineering and system software gain more opportunity to capture optical value; ANET is best positioned among listed US vendors, though its XPO approach preserves customer serviceability and may be the more commercially acceptable bridge.

Over the next 1-3 months, Huawei Connect disclosures are the relevant catalyst: a disclosed SuperPOD launch date, optics partners, or deployment scale would validate whether this is a product demonstration or an internally captive architecture. The more important 6-18 month signal is OIF convergence on lane counts, connectors and electrical interfaces. Failure to converge creates fragmented ecosystems, raising qualification costs and favoring incumbent pluggables/XPO rather than producing a rapid NPO substitution cycle.

Consensus may overstate the threat to NVDA and US networking vendors. China-restricted AI clusters already have incentives to localize interconnects, while hyperscalers outside China optimize around installed operational practices, field replacement and multi-vendor sourcing. Huawei's claimed power and latency gains remain unverified at rack scale; optical yield, thermal reliability, repair rates and total cost of ownership—not a module-level power claim—will determine adoption. Export-control escalation is a two-sided risk: it could confine Huawei's standard to China, but also accelerate a separate Chinese optical supply chain and reduce US suppliers' access to that demand pool.

Key Decisions for Investors

• No directional trade in CSCO, HPE or NVDA from this development alone; require independent evidence of hyperscaler or non-Huawei switch-vendor NPO qualification before assigning earnings impact.
• Maintain ANET as the relative long within AI networking, preferably long ANET / short CSCO over 6-12 months. ANET's system-level exposure to high-speed AI fabrics and optionality in XPO offer upside if power density becomes the binding constraint; exit or reassess if AI-networking orders decelerate for two consecutive quarters or gross-margin guidance weakens materially.
• Establish a research alert around Huawei Connect and subsequent OIF meetings: a named external NPO customer, third-party optics supplier, or standard alignment with the 6.4T interface would be a signal to revisit a short basket of merchant optical-component exposure. Without that evidence, an NPO-driven short is premature.
• Watch China AI-cluster capex and export-control actions over the next 3-6 months. A tighter restriction that forces further domestic substitution is modestly negative for US optical/networking China revenue, but does not impair the broader AI capex thesis; treat it as a regional mix risk rather than a reason to short NVDA.