Spirent Communications published a new report highlighting how AI and generative AI (genAI) technologies are driving faster optical networking port speeds using high-speed Ethernet across the telecoms and enterprise sectors.

In an interview following the report’s release, Stephen Douglas, head of market strategy for Spirent, told Mobile World Live Tier 1 operators are moving to 400G speeds as part of their IP core refresh efforts and a move to standalone (SA) 5G network launches.

He explained while some of the operators are moving to 400G ahead of SA 5G upgrades and 5G Advanced in the future, others are doing so after they’ve rolled out SA 5G.

Operators are also using 25G and 50G at “far edge sites” and 100G to 400G at mid and near edge sites.

He added operators are moving to faster port speeds to forecast or plan for capacity at the edges of their network for AI inferencing.

“A number of the telcos are starting to offer basically their network infrastructure as a service for enterprise and for sovereign AI initiatives,” he said.

Douglas noted over the past six months Spirent has seen a lot of interest by operators to design their transport infrastructure to support multi-tenant AI for enterprises, large language models and for sovereign AI.

“The large language model tenants would be hosted in their central data centres in the telco environment, but they would push inferencing out to the edges of the networks where they’re all starting to do the MEC implementations today,” he explained.

He stated with both 5G RAN and open RAN, there are more upgrades to 25G at cell site interconnects.

Douglas noted the faster speeds are driven by the demands of MIMO on both the traditional common public radio interface (CPRI) and enhanced CPRI.

He stated Spirent’s report showed large mobile operators are bringing 100G capacity to the location where centralised units sit in an open RAN architecture to perform functions such as baseband tooling.

Operators need higher capacity transport at the edge to support AI-based video demand and augmented reality.

Service providers are also focusing on the transport infrastructure for network slices, particularly for wholesale and enterprise private network offerings.

Dedicated hard networking slicing is being tested for mission-critical services with service level agreement. Flexible, on-demand soft slicing is being developed for segment routing and remote work with VPNs.

Enterprise
Douglas said large enterprises are transitioning to 400G in their data centres as part of an effort to focus on AI requirements.

Enterprise data centers are still using 100G Ethernet networks but there is growing interest in gradually shifting to 200G and 400G Ethernet for switch spine and fabric.

He noted there’s increased interest for 25G and 100G across medium and large campus networks. Wi-Fi 6 and 6E access points are driving demand for 1G at access layer but Douglas expects Wi-Fi 7 will “probably push that to five or even 10 gig requirements over the next few years”.