AI product launch: Huawei challenges enterprise optical fabric with 7.2 Tbps module

On September 10, 2026, at the China International Optoelectronic Exposition in Shenzhen, Huawei unveiled what it calls the world's first near-packaged optics (NPO) optical module capable of reaching 7.2 terabits per second, combining 36 channels of 200 gigabits each, as reported by the South China Morning Post^[1]. This is yet another signal that in the AI product launch market, the competitive race is shifting targets: from computing silicon to the fabric connecting thousands of processors together.

What actually happened in Shenzhen

The module presented by Huawei replaces copper with light to transport data between AI accelerators, reducing signal loss and power consumption compared to current systems.

Man Jiangwei, director of the company's advanced optoelectronics laboratory, told the South China Morning Post that the product has entered the industrial development phase and that mass production will arrive once the production supply chain is ready.

It is a product announcement, still distant from large-scale commercial availability: the distinction matters for those planning purchases in the coming quarters.

The real bottleneck shifts

For years, enterprise competition has focused on chips and computing power: faster GPUs, larger clusters.

With thousands of processors interconnected, the physical limits of copper become evident in bandwidth, transmission distance and energy efficiency, exactly the technical point that the South China Morning Post article highlights.

The strategic message is clear: whoever controls optical density inside the data center gains an advantage that lasts longer than that tied to a single chip.

The competitive front: Huawei versus Broadcom

The Huawei module enters direct competition with the 6.4 Tbps optical engines developed by Broadcom for co-packaged optics, the equivalent American technology aimed at the same next-generation AI clusters.

The capacity difference, 7.2 versus 6.4 Tbps, becomes an immediate sales argument. The real advantage will be played out over production volumes, price per port and ability to integrate with network switches already installed in enterprise data centers.

The market has already begun to read this race as a standardization war: whoever sets the dominant format first captures the multi-year supply contracts that will follow.

From chip to fabric: the competitive axis changes

The transition described here represents a precise shift: from chip competition to interconnection fabric competition.

The vendor with the highest optical density will capture more stable revenue than the vendor with the fastest chip, because the fabric remains embedded in the data center architecture for years, while the chip is replaced with each cycle.

This also repositions lock-in risk: enterprise customers choosing a proprietary optical fabric tie their entire future infrastructure to that standard.

Supply chain risk for European boards

For European Chief Strategy Officers, the emergence of two competing standards, one Chinese and one American, complicates the choice of suppliers for next-generation AI data centers.

• Diversifying optical component suppliers reduces exposure to a single geopolitical standard.
• Evaluating the compatibility of the chosen fabric with switches and accelerators already in the portfolio avoids future migration costs.
• Monitoring industrial production timelines, stated by Huawei as imminent, allows calibration of the investment schedule.

Technological sovereignty, already evident in European choices on models and data residency, now extends to the physical layer of infrastructure.

Optical sovereignty as a new product

Optical fabric follows the same pattern already seen with language models: the supplier's jurisdiction becomes part of the product itself.

A board choosing Chinese optical components accepts different implications, in terms of compliance and supply continuity, than a board choosing American components.

This adds another variable to the vendor map, to be read by geographic origin before even pure technical performance.

What to decide in the next 90 days

The affected functions must move on three parallel fronts, before production volumes consolidate the balance of power.

1. The Chief Strategy Officer evaluates direct partnerships with optical component manufacturers, avoiding dependence on a single supplier for the next AI cluster cycle.
2. The CFO reviews the budget line for data center interconnection, today underestimated compared to pure computing.
3. The Chief Digital Officer updates the scorecard for infrastructure vendors including optical density and geographic origin among the selection criteria.
4. The technology investor verifies which market thesis, that of the chip as moat or that of the fabric as moat, finds confirmation in the upcoming production announcements.

The decision window remains open. It narrows as soon as Huawei and Broadcom announce first volume deliveries.

The market has already moved

The Shenzhen announcement confirms a trend already visible in the AI product launch market: competitive value slips from the individual component toward the overall data center architecture.

Whoever owns the dominant optical fabric will hold a lasting contractual advantage with enterprise customers, regardless of which accelerator is chosen in any given quarter.

For boards planning AI clusters in the coming two years, the optical variable now enters among the decisive criteria, alongside compute price and chip availability.