For mid-market neocloud Verda, power availability, not low-latency connectivity, is the ultimate priority when choosing its data centers. Speaking at the recent Xcelerated Compute Show, VP Kal Aeolian explained that Verda's current workloads do not necessarily justify the low-latency connectivity that wholesale operators anticipate from the wider neocloud ecosystem.
"We're getting the megawatts where we can because the big players have already taken the majority of them," Aeolian said. "Latency is not that big of a deal with inference, or especially for training, but it might be in the future."
That presents a planning challenge for wholesale network operators such as Orange Wholesale and Virgin Media Business Wholesale (VMBW), which must anticipate connectivity demand as neoclouds enlarge their footprint across Europe wherever sufficient power is available.
Verda's view tallies with the vision of Virgin Media Business, which believes the rise of AI inference in tandem with scaling AI-centric labs and cloud will make regional and metro connectivity increasingly important. Ben Archer, VMBW's head of business development and network infrastructure sales, explained that the networking bottleneck presented by this scale-out "isn’t simply a need for more bandwidth."
"It’s whether you can get the right fiber, capacity, and route diversity to the places where power and compute are becoming available," Archer told this title. "As data center development increasingly follows power into new locations, connectivity is a key consideration, as in some places it can become the factor that determines whether that compute is genuinely usable."
Where some neoclouds like Verda are currently prioritizing power availability, the British operator is preparing its network for the more distributed traffic patterns it predicts inference will generate.
Archer explained that where AI training traditionally concentrated huge bandwidth requirements around large compute clusters, inference is instead pushing accelerated compute closer to enterprises and end users.
“Which means more regional and edge locations need low-latency, resilient connectivity … What changes is the importance of extending high-capacity connectivity from the national core into regional metros, edge data centers, and ultimately closer to enterprise locations,” Archer said, explaining that the operator is no longer focusing exclusively on big cities like London and Manchester.
“We’ve already been investing with that model in mind: aggregating traffic closer to the end user, extending high-capacity services through regional hubs and proactively connecting regional data centers.”
Virgin Media Business has been extending such connectivity in various ways, including an investment program supporting NLighten's edge data centers, a new optical connection to Kao Data's Harlow-based campus, alongside broader upgrades to its national and metro networks.
The future is Orange
Orange Wholesale faces similar infra challenges on long-haul routes across mainland Europe. In an SDxCentral interview at the Xcelerated Compute Show, head of capacity services and infrastructure Younès Stouky said emerging AI-focused facilities may require several high-capacity network paths in locations where sufficient fiber infrastructure is yet to exist.
"The challenge is always to be first to know what's coming next," Stouky added.
Orange has announced a high fiber count route across France to improve connectivity diversity between Spain, Frankfurt, and Paris, with plans to expand the approach elsewhere. Such construction can take 18 to 24 months, requiring Orange to anticipate network demand before prospective data-center customers necessarily secure their power supplies.
Of those customers, nascent neoclouds may need greater support from service providers in comparison.
"Their teams are small, so usually they come to the service provider, designing specifically what they need," the executive said.
The differences between neocloud customers could prove as important as the distinction between neoclouds and hyperscalers, along with traffic dispersion between the two camps.
Verda's Aeolian described itself as in the middle of the neoscaler spectrum, which ranged from “scrappy little startups with one small megawatt someplace to people like Nebius who are talking about multiple gigawatts.”
Orange similarly distinguished between neoclouds at different stages of development, with Stouky suggesting that managed optical networks could be especially useful for operators developing facilities with power requirements of hundreds of megawatts or more.
Virgin Media Business, meanwhile, acknowledged that unlike hyperscalers, neoclouds are not yet a dominant source of U.K. network traffic, as they might be in other areas of the world.
"Their direct impact on overall network traffic today shouldn’t be overstated, but their influence on where future capacity is likely to be required is significant," Archer told SDxCentral.
For wholesale operators, that creates an investment challenge in preparing high-capacity infrastructure and managed services for neoclouds whose locations, scale, and networking requirements may change as they expand.
For now, their diverse requirements are pushing operators to balance network scalability and operational flexibility against expected demand on the other side of the infra buildout.
Additional reporting by Paul Lipscombe.
Comments