Huawei Pitches 800G Optics and 50G PON as the Backbone for Operator AI Clusters at UBBF 2026

Huawei used its keynote at the Ultra-Broadband Forum in Vienna on October 13 to argue that operator networks now set the pace for large AI builds. Richard Liu, a Huawei vice president and president of ICT marketing and solution sales, described an upgrade path running from data centre networks through bearer networks to access networks, and hung the commercial case on a target of 100,000-card computing clusters.

Clusters at that scale need a two-tier networking architecture, Liu said, because the conventional three-tier layout multiplies switches and fibre runs. Huawei bundles three items into the answer: the two-tier design, network packet load balancing to spread traffic across links, and FlashBoot 2.0 for faster cluster start-up. Cross-border traffic growth is folded into the same plan, with submarine and terrestrial cables treated as one resource so that international links stay available to AI data centres.

On the transmission side Huawei points to 800G wavelengths and all-optical cross-connects, arranged to hold latency inside a one to three millisecond circle between computing sites. Optical switching completed within 50 milliseconds is meant to underpin availability of 99.999 percent, and wide-area lossless RDMA transmission is claimed to remove packet loss on long-haul runs between clusters.

The access layer carries 50G PON for multi-gigabit and 10G services, Wi-Fi 7 home terminals, and intelligent broadband gateways that identify traffic and schedule it. On security, Huawei describes line boards with intrinsic security and quantum encryption built in, a combination it says takes 60 percent off network construction cost.

Richard Liu framed the argument in one sentence: “Networks are the foundation of operators. By building AI-centric new generation UBB network, together we will bring intelligence to every person, every home, and every enterprise.”

Independent market data explains the timing. Dell’Oro Group’s optical transport report, published in August, put year-over-year growth in that segment at 15 percent in the second quarter of 2026, with data centre interconnect revenue from IPoDWDM ZR/ZR+ pluggables and WDM systems up 45 percent. Cloud providers accounted for 34 percent of optical transport revenue, and Huawei, Ciena, Nokia and Cisco held the top four places over the trailing four quarters.

The forum, now in its twelfth year, ran alongside Network X in the same city, with Austria’s telecom regulator among the opening speakers.

Supply note: For equipment buyers, the timing of these AI-centric designs shows up first in optical stock. Coherent transponders rated at 800G are on the shelf now in both fixed and pluggable form, alongside 100G and 200G QSFP28 and QSFP56 modules that swap into existing OLT and OTN shelves without a redesign, and cross-referenced equivalents for the cases where an OEM part sits on allocation. In-stock items carry a two to four working day lead time out of Hong Kong, with cross-reference sourcing adding about a week. Bench history on returned 50G PON and 800G units points at contaminated fibre end-faces and mismatched optical budgets far more often than failed lasers, so each unit leaves with a cleaning kit and a recorded insertion-loss figure. Capacity-expansion orders usually arrive as mixed pallets built to a given rack layout and cable run.