KDDI preps for AI era with disaggregated routers

• Japanese operator KDDI has been trialling and deploying disaggregated router technology for more than four years to achieve greater flexibility and scale in its data transport network
• It has already deployed distributed disaggregated backbone router (DDBR) technology at its internet gateway peering points
• Now it is ready to deploy clusters of DDBR systems to cope with AI-fuelled data traffic volume growth, some of which will come from its own AI datacentre facilities
• The move comes as KDDI also accelerates its Open RAN rollout plans 

In a reminder that network disaggregation is not limited to the challenges associated with evolution of the radio access network (RAN), KDDI has unveiled plans to deploy clusters of disaggregated backbone routers that will help it to “accommodate the increasing traffic demands of the AI era,” the Japanese telco has announced.

KDDI has been at the forefront of disaggregated routing system developments for more than four years already, having played a key role at the Telecom Infra Project (TIP), where its representatives chaired the Disaggregated Open Routers (DOR) subgroup in the TIP Open Optical Packet Transport (OOPT) project group and, in its own words, was “active in [the] technical development of carrier-grade white box routers” – see KDDI reminds us that open networking is about more than just the RAN.

Following multiple successful trials and tests, KDDI began deploying distributed disaggregated backbone router (DDBR) systems in its live network at its internet gateway peering points in June 2023, noting at the time that by using DDBR technology it would be able to reduce its power consumption by about 46% and rack space by about 40% compared to the traditional routers in KDDI production networks. The DDBR systems comprised the cloud-native DriveNets Network Cloud software stack (including its network operating system, or NOS), line cards from Delta Electronics (40x100 Gigabit Ethernet) and merchant silicon (Jericho2) from Broadcom. 

Now, following successful trials, KDDI is to deploy clusters of DDBR systems as part of what it calls its “Scale-Out Network” that supports its fixed and mobile traffic: By the end of the current financial year, which ends in March, the operator aims to have deployed DDBR clusters at four commercial backbone network hubs, while a full nationwide rollout is targeted by the end of the financial year that runs April 2026 to March 2027. 

It’s also worth noting that KDDI is targeting late calendar year 2026 for the launch of AI services from a new datacentre being developed on the site of a former Sharp plant. When, in June 2024, it first announced its plan to build the facility, KDDI claimed it would be “the largest AI datacentre in Asia”.

The deployment of DDBR clusters will, the operator hopes, give KDDI the flexibility it needs to deal with the hard-to-predict growth in data traffic running over its networks. As KDDI explains: “Unlike traditional chassis-based routers, which integrate hardware and software within a single system [see below left],” and which require entire chassis upgrades to add capacity, “the DDBR separates the two [see below right]. This approach eliminates slot constraints found in conventional chassis designs, allowing for rapid and flexible scaling. For instance, by adding line cards to connect additional cables, the DDBR can flexibly scale out to accommodate increasing traffic.”  

This approach also gives KDDI greater flexibility with regards to the size and configuration of the clusters. KDDI notes that the “DDBR cluster allows hardware to be added as needed, ensuring flexible capacity expansion to accommodate traffic growth. Unlike traditional large-scale routers, which rely on a single system, the DDBR cluster enables traffic distribution across multiple smaller routers. This flexibility allows larger DDBR clusters to be reconfigured into smaller clusters based on traffic volume, significantly improving redundancy and enhancing overall network reliability.”

It also helps with KDDI’s network planning and capex management. “The same architecture and hardware can be used across peering, core, and edge network domains. This allows for standardised spare parts, shared testing and operational knowledge, and reduced hardware costs, leading to more efficient network management. Additionally, because the system is vendor-agnostic, it can integrate hardware and software from multiple vendors, providing the flexibility to adopt new technologies and features as needed,” noted the operator. 

KDDI plans to show off its DDBR cluster architecture during the upcoming MWC25 show in Barcelona (at its stand in hall 4). 

The shift towards disaggregated data networking systems is part of a broader technology and operational revamp at KDDI, which is also embracing Open RAN in its mobile network infrastructure as part of its Vision 2030 strategy that the operator unveiled in 2022. KDDI announced in October 2024 that, following years of pilot deployments, it is ready to ramp up its Open RAN-enabled rollout in 2025 and had selected Samsung Networks as its main technology partner, with Hewlett Packard Enterprise (HPE), Intel and Red Hat also part of the technology mix. 

And all of these developments will soon be overseen by a new CEO as, after seven years at the helm, 63-year-old Makoto Takahashi is taking on the role of chairman from 1 April and will be succeeded as CEO by Hiromichi Matsuda (53), currently chief digital officer (CDO), the operator announced earlier this month. The current chairman, 67-year-old Takashi Tanaka, is taking on a role as an advisor to the KDDI board.   

- Ray Le Maistre, Editorial Director, TelecomTV