Networks not ready for the challenges of AI traffic • The Register

AI is reshaping the calls for on community infrastructure, and plenty of organizations aren’t ready – together with a few of the so-called neocloud suppliers providing AI providers.

A research by analyst biz Omdia finds that many rent-a-GPU suppliers have scaled up their compute infrastructure to deal with AI workloads, however their networking infrastructure is changing into a crucial constraint.

It warns enterprise prospects to scrutinize potential suppliers past their uncooked compute capability when contemplating AI compute providers.

Neocloud operators, or GPU-as-a-service suppliers, sprang as much as make the most of the large demand for compute utilizing GPU accelerators for AI. Many rely hyperscalers reminiscent of Microsoft amongst their prospects, in addition to enterprise.

Which means AI efficiency more and more is determined by their means to course of and transfer information securely throughout distributed environments and geographies. Nonetheless, the networking capabilities of various neoclouds fluctuate from rudimentary to superior, relying partially on their origins, Omdia says.

Some, reminiscent of CoreWeave, began life as cryptocurrency mining operations, whereas others, reminiscent of Gcore, beforehand centered on content material distribution or webhosting. 

Due to this, neocloud networking technique is in flux globally, it says, with many speeding to associate, purchase, or construct infrastructure as their dependency on networking will increase.

“Community infrastructure will make or break neoclouds,” warns Omdia Telco B2B Analysis Director Camille Mendler. “Low latency, resilient and safe connectivity from spine to edge is desk stakes for achievement, not least as a result of sovereignty spans the place AI workloads transfer,” she added.

International community supplier Lumen is leaping on the identical bandwagon. CEO Kate Johnson issued an open letter to enterprise chiefs in every single place asking if their networks are AI-ready and pushing upgrades to assist coming AI purposes, as properly she may.

Networking has historically been within the background, like plumbing, Johnson claims. “However in an AI-driven enterprise, the community is extra just like the nervous system. It controls and coordinates. It determines how briskly you may transfer and whether or not your AI investments produce worth,” she says.

AI methods do not function in a single single location, however contain fixed information motion between clouds, datacenters, and edge endpoints, and so networks have to be adaptable and in a position to scale dynamically.

“The brand new company workforce is comprised of AI brokers and bots. They’re proliferating quickly, working repeatedly, insatiably consuming and producing information and dynamically interacting with different brokers, bots and people,” Johnson says within the letter.

“And regardless of the early days of AI adoption in most companies, right now, greater than 50 p.c of web visitors is created by these autonomous staff,” she claims.

This declare comes from Imperva’s 2025 Bad Bot Report, which states that automated visitors has now surpassed human exercise, accounting for 51 p.c of all web transmissions.

“To assist the courageous new world of AI, networks must be utterly adaptable, programmable and consumption-based, identical to cloud,” Johnson states, earlier than exhorting enterprise chiefs to “be sure your community helps the long run you are constructing.” ®