Research from network vendor Ciena found operators underprepared for the impact AI workloads will have on optical and IP networks.

In a global survey of communications service providers (CSPs), only 16% of operators reported their optical networks as “very ready” for AI, while nearly one-third (29%) believe AI will contribute more than half of all long-haul traffic over the next three years.

The study, which was carried out with analyst firm Heavy Reading, assessed the impacts of AI applications and traffic growth on metro and long-haul networks run by fixed, mobile, and converged network operators and cable operators.

Around one-fifth of CSPs (18%) expect AI to contribute more than half of their total metro network traffic, while nearly half (49%) believe AI will exceed 30% of metro traffic.

There were higher expectations for AI concerning long-haul traffic, with 52% of CSPs expecting AI to exceed 30% of traffic by 2028.

Ciena reported 39% of networks are positioned for the AI boom, with most of the necessary infrastructure in place but some upgrades still required. According to operators, the biggest obstacles to reinforcement are capex limits and unclear commercial strategies – each cited by 38% of participants – followed by the complexity of managing multilayer networks at 32%.

Nevertheless, CSPs see a revenue opportunity: around three-quarters (74%) expect enterprise customers, rather than hyperscalers, to drive most incremental AI traffic, creating fresh demand for managed wavelength services at 100 Gb/s, 400 Gb/s, and 800 Gb/s.

Half of those surveyed rank these high-bandwidth wave offerings as the fastest-growing service category tied to AI, coming out ahead of unused dark fiber.

“This research highlights the rapid rise of AI applications – from large-scale models to cloud AI services and edge-to-core workflows – that are set to become major drivers of both local and long-haul network traffic,” Sterling Perrin, senior principal analyst at Heavy Reading, noted. “For metro networks, where AI will compete with video, web, and IoT traffic, the projected growth is striking. With AI expected to take an even larger share of long-haul capacity within three years, it’s clear that AI data flows, including those used for training and inference, will put unprecedented demands on CSP networks.”

Supporting views

The research reflects similar findings on the impact of AI on networking. A recent report by Dell’Oro Group found that AI led to a 51% increase in global data center capex in 2024, hitting $455 billion in investment.

While Ciena’s research expects enterprise customers to be a major force in the future, hyperscalers remain today’s biggest drivers. More than half of last year’s investment was driven by custom accelerators from the likes of Amazon Web Services (AWS), Microsoft, and Google Cloud, with their AI training workloads fueling most of the growth.

Real estate consulting firm JLL’s Data Center Project Development and Services, meanwhile, expects around 40% of total data center workloads to be AI-related by 2030.

650 Group argued this AI demand will push Ethernet, fueled by growing adoption of 800 Gb/s (800G) and eventually 1.6 Tb/s (1.6T) system capacity, past InfiniBand as “the dominant technology for scale-out” by the end of this year, generating more than $8 billion in revenues.

Adding to AI workloads will be the rise of AI agents, which are AI models that perform tasks autonomously on behalf of enterprises.

Cisco recently suggested to SDxCentral that the impact of numerous AI agent deployments could strain network bandwidths equivalent to “80 billion” users.