- Equinix exec Arun Dev said connectivity is an often-overlooked piece in the AI puzzle
- Equinix and Zayo have created a network blueprint for AI workload
- Neoclouds and large enterprises are among the target audience
Neocloud providers and large enterprises are rushing into AI, looking to get workloads up and running as fast as possible. But while they’ve been busy focusing on sourcing power, space and compute, there’s one thing they haven’t been thinking enough about: the network.
“Everybody overlooks the connectivity piece,” Arun Dev, Equinix’s VP of Digital Interconnection, told Fierce. It’s well and good to build a massive data center in west Texas or Louisiana, until you realize that there’s not a whole heck of a lot of connectivity in that region, he said.
“Most folks don’t know what to do,” he said. “It was all about ‘where are the GPUs?’ So you went a found a whole bunch of GPUs in Louisiana. Then a lot of these new providers are sitting and saying, ‘hey, how do I connect to these cloud workloads in US-East?’”
There’s more to it than just hooking up a data center to any old fiber, he added. Not only do neoclouds and enterprises need to figure out how to get from the data center to users, but they also have to figure out how to connect to the cloud along the way. The kind of network connectivity needed also depends somewhat on whether someone is running a training or an inference workload.
After fielding requests for help from their respective customers, Equinix and Zayo decided to team up to provide a roadmap in the form of the pair’s new AI Infrastructure Blueprint.
The Blueprint is basically a network infrastructure cheat sheet, outlining what kind of network components (eg., metro vs long-haul fiber, dedicated vs public internet, and interconnect and training nodes) are needed for consumer and enterprise training and inferencing workloads. It also provides a list of key considerations for AI backbone networks and examples of both network architecture and topology.
It might seem like Equinix and Zayo are playing to a tiny niche here, but depending on which count you believe, there are anywhere from 54 to 190 neocloud players alone. Think the likes of CoreWeave, Vultr, Crusoe, NScale, Lambda, Nebius, and more.
Plus, despite a recent slowdown in uptake among some U.S. companies, AI adoption broadly continues to rise. Dev said the blueprint will be most useful to “advanced AI enterprises.”
Equinix and Zayo aren’t the only ones looking to make AI adoption easier for enterprises. Lumen also recently expanded its network-as-a-service offering to serve more off-net workloads and prepare enterprises for the “AI economy.” You can read more about that here.