1
While AI-related traffic is affecting multiple network segments, Brodie Gage, senior vice president of global products and supply chain at Ciena, said the most immediate pressure is emerging in metro networks, particularly DCI environments and AI multi-cloud access networks
In sum – what to know:
Scale-across – AI training is increasingly distributed across multiple data centers, campuses, and regions, requiring new network architectures that function as an extension of the data center fabric.
Metro pressure – The most immediate network pressure is building in metro DCI and AI multi-cloud access networks as AI workloads expand beyond individual facilities.
Fiber limits – While 800G and 1.6T optical technologies can extract more capacity from existing infrastructure, Ciena says they are not sufficient to meet AI-driven connectivity growth on their own.
The rapid growth of artificial intelligence is reshaping network architectures, traffic patterns, and infrastructure requirements, creating new challenges for optical networking providers and operators alike.
In an interview with RCR Wireless News, Brodie Gage, senior vice president of global products and supply chain at Ciena, said that AI workloads differ fundamentally from traditional cloud applications because they are significantly more complex, more bandwidth-intensive, and require highly reliable, low-latency connectivity between large clusters of GPUs.
One of the most important changes is the emergence of what Gage describes as “scale-across” architectures. Rather than concentrating AI training within a single facility, organizations are increasingly distributing GPU clusters across multiple data centers, campuses and even regions.
“Scale-across extends AI training beyond a single data center, enabling one AI model to be trained across GPUs located in different campuses or even regions,” Gage said. The result is a single logical AI training environment built on physically distributed infrastructure.
This shift is being driven in part by power density constraints, which are making it more difficult to deploy all AI infrastructure within a single location. As a result, network connectivity has become a critical component of AI infrastructure deployment.
The architectural transition is creating new demands on optical networks. According to Gage, AI workloads require an order of magnitude more capacity than traditional metro data center interconnect (DCI) networks. To address these requirements, Ciena has introduced technologies including 800Gb/s C&L-band coherent pluggables and hyper-rail photonics designed to increase connectivity, fiber density and network efficiency.
AI workloads also require higher data transmission rates, lower latency, and greater energy efficiency. At the same time, operators are under pressure to deploy infrastructure rapidly because AI facilities only begin generating value once connectivity is fully operational.
While AI-related traffic is affecting multiple network segments, Gage said the most immediate pressure is emerging in metro networks, particularly DCI environments and AI multi-cloud access networks.
Initially, demand was concentrated within specialized AI clusters located in large data center campuses. As AI training expanded across distributed sites, pressure began spreading into metro and long-haul DCI routes. Hyperscalers and neoscalers are now deploying routes with up to hundreds of fiber pairs operating in parallel to deliver tens of petabits per second.
The rise of multimodal AI is also changing traffic dynamics. Gage noted that text-based AI had a more limited impact on network demand, while multimodal applications are creating substantially greater bandwidth requirements.
To meet growing demand, operators are adopting higher-capacity optical technologies. According to Gage, 800G and 1.6T coherent optical technologies can extract 30% to 50% more capacity from existing infrastructure, helping extend the life of deployed fiber assets and absorb near-term AI traffic growth.
However, he cautioned that AI is driving traffic increases of 10 to 100 times or more on certain routes. As a result, optical upgrades alone are not enough. “While new 800G and 1.6T technologies are somewhat slowing the need for new fiber, they do not come close to meeting connectivity requirements needed,” Gage said. “Fiber expansion or a different architectural approach is critical.”
Looking ahead, Gage said there remains significant room for innovation in coherent optics and photonic systems. He pointed to developments such as hyper-rail photonic line systems, full-spectrum transponders, co-packaged optics, and hollow-core fiber as examples of technologies being evaluated by hyperscalers.
As AI infrastructure continues to expand, Gage expects optical innovation and architectural evolution to advance together, helping networks support increasingly distributed, bandwidth-intensive AI environments.
The interview with Ciena’s Brodie Gage is part of a recent report published by RCR Wireless News and RCRTech, titled Scaling Optical Networks for the Hyperscale and AI Era, which can be accessed by clicking here.
