AI is causing a real impact on networks, forcing data center and network owners to re-evaluate their current technology choices and how they will be positioned to face an expected onslaught of AI-generated networking demands. The actual mass of that onslaught remains nebulous, but most agree it will be substantial and will require significant financial investments to support.

Most of this initial AI-related traffic generation is happening within data center fabrics. These are the interconnected links that tie together the thousands of compute boxes running AI workloads within a data center location. This traffic demand is what is driving significant investments into building out new and updating legacy data center locations.

“Certainly, AI is sort of the overarching theme behind a lot of the things that are driving capacity in customer networks, whether it's hyperscalers or service providers, and ultimately enterprise networks as well,” Bill Gartner, SVP and GM for optical systems at Cisco’s Optics Group, said during a fireside chat session at the recent OFC 2025 Conference.

Dell’Oro Group in a recent report found that global data center capex increased 51% in 2024, hitting $455 billion in investment last year. The research firm noted that custom accelerators from hyperscalers like Amazon Web Services (AWS), Microsoft, and Google Cloud accounted for more than half of that total investment, and most of last year’s growth was fed by those hyperscalers bolstering their support for AI training workloads.

Matt Landek, who leads real estate consulting firm JLL’s Data Center Project Development and Services, said that his firm is currently quoting only around 15% of total data center workloads are AI-related, but that they expect that number to hit around 40% by 2030. Analyst firm LightCounting last year predicted growing AI use will result in a doubling of sales for Ethernet optical transceivers used in AI clusters.

AI driving inter-data center networking

There is also growing demand to bolster inter-data center networking connections to help support AI-related data inferencing outside of those main data center locations. This is typically targeted at improving the performance of optical network connections between large data centers in different regions or to edge locations that are closer to end users where most think data inferencing will take place.

These investments are typically more involved than intra-data center updates as they include more complicated equipment in less controlled environments, but there are considerable efforts in place to boost these inter-data center connections. Large players like Huawei, Ciena, and Nokia hold sway in terms of market share, with each also pushing out higher-speed options using advanced fiber optic and transceiver technology that can support speeds up to 1.6 Tb/s.

Dell’Oro Group noted that the optical transport market ended last year “on a strong note, growing approximately 45% quarter-over-quarter” during the final three months of the year.

“There is light at the end of this tunnel,” Dell’Oro Group VP Jimmy Yu wrote. “The optical transport equipment market entered a period of revenue contraction in late 2023 and looks to be exiting it in early 2025. The strong fourth quarter growth rate in 2024 is a great indication that supply and demand for optical transport equipment is near equilibrium, meaning the equipment glut is over.”

Those market leaders are jumping at this new opportunity.

“I think we are in an unprecedented industry disruption, and what we have seen over the last couple of years is just a commitment to build compute, and while networking had a little bit of a back seat, the conferences over the last 12 months have completely changed that,” Jürgen Hatheier, CTO for Ciena’s International business, told SDxCentral. “Networking has become one of the sexy things again, where hyperscalers have to pay attention not only to the fastest Nvidia GPUs and the latest Google TPUs… but they need to build an ecosystem. And while it has been built a little bit by brute force … within the data center, just more capacity, more capacity, there is now a lot more sophistication in how it is being built.”

Hatheier’s “sexy” comment was echoed by Nokia’s Manish Gulyani, who said, “IP used to be the sexy thing and optical was, that's fine. I think it's kind of reversed. IP is like it's all nothing new happening in IP, and optical every day there's some new, cool innovation, because I think with AI, optical is on fire.”

However, the size of that fire will be tempered by financial fuel.

“All that compute is generating so much data that will now need to be consumed, that will now need to leave all these mega-, hyper-data centers to be consumable for enterprises and consumers alike,” Hatheier said. “Training costs money, but inference and reasoning is what ultimately will generate money for the hyperscalers.”

Rob Shore, who previously worked at Infinera and is now part of portfolio marketing for optical networks at Nokia, agreed, noting that this balance will require further innovation.

“I think there's a whole new breed of optics that are going to be needed,” Shore said. “In addition to having trouble with the power grids, data center operators are having trouble fitting everything into a single building, so what they're having to do is break up a single data center into a bunch of smaller data centers that they want to act like a single data center, but those could be as far as 20 kilometers apart. So, how do I interconnect those with low-cost optics, low latency optics?”

And it’s not just cost. Matt Rehder, VP of core networking at AWS, explained to SDxCentral that the hyperscaler is really more interested in stable connectivity rather than the highest performing connection to support the initial rush around AI and generative AI.

“A lot of the generative AI workloads, they're just more sensitive to failures. They're running at higher tolerances, and we have protocols, we have everything else in place to make the network as resilient as possible and react to failures,” Rehder said. “But if you can fundamentally have your network fail less because your devices and links are just significantly more reliable, it's a much easier foundation to build on.”

Network operators sense a networking opportunity

This AI-infused opportunity is driving network operators to re-evaluate their own business models that have traditionally been mostly focused on basic connectivity.

Lumen Technologies, for instance, is betting big on an AI-infused networking future that will demand significantly higher throughput and lower latency than what is being provided today and will drive what CTO Dave Ward previously told SDxCentral will be “the largest expansion of the internet in our lifetime.”

Lumen’s investments are geared toward supporting what the vendor said is a standing $5 billion in aggregate sales it has already booked and the opportunity toward an additional $7 billion in sales pipeline.

Lumen CEO Kate Johnson told investors during an earnings call last year that the $5 billion in aggregate sales is from “hyperscalers, it’s social platforms, it’s huge technology companies, it’s a cloud company.”

Johnson said she expects the next “tranche of demand” to come from the “AI model inference phase, probably with forward-thinking enterprises who see AI as a way to transform their businesses. Think financial services, health care, and retailers to start.”

Ward explained that this AI inference push will also spill over into hybrid architectures that will see data held on-premises and using inference models associated with trained models in the cloud. “Massive amounts of bandwidth and super tight latency requirements for that,” Ward said.

“What this means for us as a connectivity partner is we fully plan on building an AI fabric between these locations and major data centers that are of the right power, size, scale, to host GPUs and the AI workloads and create a specialized connectivity fabric just for that purpose,” Ward said of this effort. “So much fiber and so many waves are required, and that segment has really become a different segment than some of the other cloud economic segments that are coming in that we can construct just to that.”

Telecom operators are also in a position to take advantage of this AI-related networking surge due to their diverse geographic footprint.

Juniper Networks CEO Rami Rahim told SDxCentral that the inferencing opportunity can be targeted at telecom operators that control prime real estate within the AI ecosystem.

“Connectivity alone is not enough to pay the bills, so one of the most important emerging models is the understanding that inferencing is not all going to happen in centralized locations. It's, in fact, going to happen closer to where the data resides,” Rahim explained. “In some cases, it might make sense for large enterprise, large financial organizations, and banks to do that inferencing on-premises. But I also think there's a very large opportunity for telcos to do that inferencing where they have beach-front property at the edge that connects directly to their enterprise customers.”

Telecom CEOs have been discussing this opportunity.

Verizon recently signed a deal with Nvidia to power enterprise AI services and digital transformation efforts running over the carrier’s 5G private network and mobile edge compute (MEC) infrastructure. Adam Koeppe, SVP of technology planning at Verizon, in a recent interview with SDxCentral, touted the carrier’s architecture and its ability to support AI-derived use cases.

“I think it’s really important to look at what these capabilities exist within the architecture, and how can AI, true AI, augment things that are already being done or create things that are brand new,” Koeppe said. “Where I see our evolution occurring is when you have an advanced cloud platform, as we do, you have an orchestration layer on top that we already have, and you then find ways to incorporate new AI capabilities on top of that. That’s going to allow your engineers and your operators to interface differently.”

T-Mobile US is also working to turn its cell site locations into edge data centers or cloud connections that can in turn better serve AI workloads.

“These kinds of AI workloads will increasingly demand that processing happens on the device, which we’re starting to see but there are obvious limitations there, or on a cloud near the device, and that’s a business opportunity that we see in the future,” T-Mobile US CEO Mike Sievert said.

Cisco’s Gartner noted that these opportunities are set to drive further investments.

“AI could be a stimulant for service provider customers that either begin to host AI applications for enterprise customers or as the enterprise market starts to build out [on-premises] solutions,” Gartner said. “We can start to see service providers having to build out more capacity as well.”