- AWS joined Microsoft and Meta in inking a network transport deal with Lumen to facilitate AI growth
- Flexential's CEO has warned network could become the next big bottleneck
- But Lumen says it has plenty of bandwidth to go around and plans to build more
Hyperscalers are buying what Lumen CEO Kate Johnson is selling when it comes to network transport for artificial intelligence (AI). Amazon Web Services (AWS) just became the third cloud titan to ink an AI-specific deal with the telecom company, following the likes of Microsoft and Meta. But what do these deals mean for other businesses that need network bandwidth for the same?
In an interview with Fierce this week, Flexential CEO Chris Downie reiterated his warning that the network could soon replace GPU supply as the next AI bottleneck. Downie noted that the scale of the network being deployed inside data centers – like those operated by Flexential – has increased “to a level we’ve never seen before” and that the data flowing within these facilities will need to be transported elsewhere.
“So, we’re expecting there to be increasing demands on the network,” he said. Couple that with the trend of data centers moving into secondary and tertiary markets in search of available power and the problem becomes one of deploying that resource to locations where it was never expected to scale fast enough to meet that rising tide, he added.
“Some look at it as a limitless resource, but I think with some of these partnerships being made it seems like there’s some preferential positioning that might make that a harder resource for others to secure at same level of scale and the same level of geographic reach,” Downie said.
But will that actually be the case? We took the question to Lumen. A company spokesperson told us Lumen has significant existing capacity that is “underused,” adding it expects to have capacity available for the “foreseeable future.”
The representative noted Lumen has also created a new division to continue expanding its network and it plans to “install new fiber routes on customers’ behalf so they can further expand their fiber capacity in the future.” To that end, it has struck a deal with Corning to ensure it has enough fiber cable available to completed these projects.
“This fiber-dense cable will more than double Lumen’s U.S. intercity fiber miles, offering significant capacity to major cloud data centers racing to stay ahead of AI workloads and high bandwidth applications fueled by massive amounts of data,” the rep stated.
Lumen has homed in on hyperscalers as a key growth opportunity, with Johnson recently telling Fierce on The Five Nine podcast it has already closed $5 billion worth of hyperscale networking deals and has another $7 billion in the pipeline. The latter tranche includes not only big tech companies but also enterprises who understand that “AI needs data, data needs data centers, data centers need to be connected.”
A hyperscale concern?
But is this kind of network bandwidth just a hyperscale concern? Or should enterprises be hustling to lock in their lanes on the data superhighway with Lumen and peers like Zayo and Windstream as well? It depends, analysts said.
“My view is that the enterprise demand and need for bandwidth for GenAI will be different from the hyperscalers," neXt Curve founder Leonard Lee told Fierce. Will hyperscalers crowd enterprises out? 'Possibly, but it would heavily depend on the future demand for distributed GenAI computing and how enterprises end up deploying."
Similarly, IDC VP Dave McCarthy noted that there’s a need to connect more than just data centers.
“I firmly believe that success in AI is going to require more than hyperscale data centers,” he wrote on LinkedIn. “The interconnection to private datacenters and edge locations will be necessary to deliver on AI inference in a performant and secure manner.”
Dell'Oro Group's Sameh Boujelbene told Fierce that while hyperscalers are indeed expected to generate a wave of AI-related backbone bandwidth demand, it only makes sense for them to ensure that enterprises can catch that wave with them.
"I think hyperscalers would need to make sure there is enough bandwidth to go around as enterprises will be key to monetize those AI investments," she concluded.