3
Bruno Berti, SVP of global product management at NTT Global Data Centers, told RCR that optical technology by itself will not remove the need for additional fiber as AI traffic scales
In sum – what to know
Optics extend fiber – Higher-speed optics can extend existing fiber, but AI traffic growth will still require targeted fiber expansion.
AI reshapes networks – AI workloads are driving flatter, high-bandwidth architectures, greater optical connectivity and tighter integration with physical infrastructure.
Fiber moves inward – Optical networking is taking a larger role inside AI facilities and across campuses as networks scale with compute.
AI traffic is changing the role of optical networking inside and between data centers, with higher-speed optics, greater fiber density, and tighter integration between networking and physical infrastructure becoming increasingly important, according to Bruno Berti, SVP of global product management at NTT Global Data Centers.
Berti told RCR Wireless News that traditional cloud traffic is relatively predictable, with traffic flowing between users and applications, moderate latency sensitivity and steady bandwidth growth. AI workloads are different, with training and large-scale inference generating enormous volumes of traffic inside data centers as GPUs communicate with each other. “In some cases, microseconds matter,” Berti said.
That is increasing the requirements placed on networks, which Berti said need to become “flatter, more deterministic, and far more resilient.” He pointed to higher-capacity fabrics and advanced optical interconnects as part of that shift, alongside closer integration between compute, storage and networking.
From a data-center perspective, Berti said the challenge extends beyond networking. Power density, cooling, fiber density, fiber routing, campus architecture, and connectivity to edge and cloud ecosystems all need to work together.
Higher-speed optics can extend the life of existing fiber, particularly at the data-center and metro levels, but Berti said optical technology by itself will not remove the need for additional fiber as AI traffic scales. He specifically addressed current higher-speed technologies including 400G, 800G, and 1.6T.
“Higher-speed optics significantly extend the life of existing fiber, especially at the data-center and metro level. But as AI traffic scales, optics alone won’t eliminate the need for new fiber — the winning model is smarter optics plus thoughtful, targeted fiber expansion,” he said.
Berti said network infrastructure is largely keeping pace with compute expansion, but this depends on deliberate planning and scaling alongside compute. In AI environments, higher-speed optics, flatter architectures and tighter integration inside data centers are helping networks keep pace.
He added that network problems can emerge when planning for the network lags behind compute planning, particularly because AI workloads are sensitive to latency, bandwidth, and packet loss within GPU clusters and between connected facilities.
“What we’re seeing is a shift from networks being treated as plumbing to being treated as core AI infrastructure. When compute, networking, power, cooling, and physical layout are designed together, scaling works. When they aren’t, bottlenecks appear,” the executive said.
Berti also said network requirements will continue to evolve as workloads shift from AI training toward AI inference and as agent AI takes over more infrastructure demand.
Architecture is evolving toward flatter, high-bandwidth fabrics inside data centers. Berti said large-scale AI training depends on constant, synchronized communication between GPUs, creating massive east-west traffic and favoring architectures that minimize hops, reduce latency, and deliver extremely predictable performance.
Optical networking is taking a larger role deeper inside AI facilities and across campuses. Berti said higher-speed optics help move data between clusters and data centers while keeping latency and energy use under control.
Looking further ahead, Berti said emerging photonic approaches, including initiatives such as IOWN, point to a future where optical technologies play an even deeper role in reducing latency and improving energy efficiency across large-scale AI environments. “These are longer-tern innovations, but they reflect the direction the industry is moving as AI traffic continues to scale.”
Network design is also becoming more closely integrated with the physical data-center environment. Berti said power density, cooling strategy, rack layout, and fiber routing all need to be engineered together, while more space is being dedicated to bringing additional fiber and conduits into data centers.
At the same time, Berti said modularity and scalability are becoming more important because AI requirements are evolving rapidly. Networks need to scale from hundreds of gigabits to terabits without disruptive redesigns, he added.
