Neolclouds have yet to build networking strong enough to handle AI workloads, according to recent research by Omdia, a Channel Dive sister company. As soaring compute demand hits connectivity constraints, partners have an opportunity to help these GPU-specialists build out their networks.

AI performance depends on the ability to process and move data securely across distributed environments and geographies. While the Omdia report, which analyzed 50 neocloud providers, targets network operators and large enterprises, the findings reveal infrastructure gaps that partners with networking, interconnection and sovereignty expertise can surmount.

“Network infrastructure will make or break neoclouds,” said Camille Mendler, Omdia research director, telco B2B. “Low-latency, resilient and secure connectivity — from backbone to edge — is table stakes for success, not least because sovereignty spans where AI workloads move.”

Connectivity hasn’t impeded neocloud revenues, which more than tripled year over year to $9 billion during the final three months of 2025, according to Synergy Research Group. It also hasn’t slowed investments in the GPU clouds, as evidenced by trading firm Jane Street’s agreement earlier this month to buy $6 billion in CoreWeave services and fork over $1 billion in equity to the neocloud.  

Meanwhile, connectivity provider Zayo secured a deal with an unnamed “leading global AI infrastructure partner” to build 8,000 route miles of new and overbuild fiber miles across high-demand corridors. The contract represents the largest investment of its kind in the company’s history, according to Zayo.

“We’re not building speculatively, we’re actively building markets where we can already see that pressure taking hold and where demand is clearly headed,” Bill Long, chief product and strategy officer at Zayo, said in the announcement. “We’re doing it at a scale that not only supports these initial commitments but puts additional capacity in place ahead of the next wave of AI demand.”

Partner opportunities

The scale of the spending suggests partners have more to gain from AI than just compute resale, especially as network capacity, interconnection and performance become difficult to separate from AI products.

Cloud and data center interconnect capacity stands out as the area in most urgent need of service provider support, according to Omdia. Campus and branch sites are the next two in line. From there, distributed inferencing, model synchronization and sovereignty requirements are reshaping what customers need from networks. Indeed, Omdia’s March traffic update noted that net new AI traffic through 2027 will favor upstream and B2B use cases. Training, retraining and distributed inferencing rank as among the fastest-growing segments.

The trend favors partners that do more than broker circuits.

Omdia’s neocloud analysis encourages service providers to position network design and control as managed services, including routing setup and traffic engineering. It also urges providers to bundle services tied to network identity, address ownership and prefix acquisition, combine IP transit with direct cloud interconnects, and make sovereignty part of the value proposition. Omdia adds that in-country operators can become trusted partners when customers need networks that meet government and regulatory requirements.

Many neoclouds remain immature on the networking side and not all will survive or thrive as businesses, according to Omdia. Networking lapses stem from a neocloud gold rush, Mendler said. Her advice to communications service providers is to “see the skull beneath the skin” and “anticipate networking constraints, likely technology debt, and emerging needs to transport GPU workloads.”

In a separate look of more than four dozen neoclouds, Omdia found that 43% were seeking network engineers and security specialists. More than half did not use internet peering exchanges. Another 46% controlled only small IPv4 address blocks and one in five relied on a single IP transit provider. 

Partners can be instrumental in identifying the right approaches to address network issues, as accountability for uptime, security and sovereignty deserves close review. Omdia warned that neoclouds relying on a single best-effort IP transit provider may not be able to control geographic traffic routing well enough to support sovereignty ambitions. Put another way, a flashy AI factory can still have a weak undercarriage.

“Read the signals,” Mendler wrote. “Don’t let high valuations obscure the realities of startup survival rates. Besides funding, indicators of maturity include AS numbers, originated IP prefixes, and IX presence.”