Loading page…
The proposed pilot would balance electricity availability against speed, security and cost. It remains an exploratory effort, not a deployed service.
Listen to this story
The collaboration would make network routing a lever for managing data-center electricity demand, rather than relying solely on adding power capacity. In an October 29 announcement, Lumen and Emerald AI said they are exploring a pilot to shift AI inference requests among locations based on available electricity. Emerald’s Conductor would work with Lumen’s programmable fiber and edge sites, while routing also accounts for latency, security, cost, and performance. The proposal targets rare grid-stress periods; no pilot results or finished service have been reported.
Emerald founder Varun Sivaram said the intended use is to ease a data center’s electricity demand during rare grid-stress periods while continuing to serve customers.
The proposed routing would extend Emerald’s power-aware technology beyond decisions made inside a data center.
Lumen’s Jim Fowler characterized the network as a potential active participant in deciding where computing happens.
Lumen and Emerald AI are exploring a pilot that would send AI requests to locations with available electricity, making power supply part of the routing decision. The collaboration would combine Emerald’s energy intelligence with Lumen’s network and edge infrastructure, aiming to keep AI services running while giving data centers more flexibility when power is constrained.
Lumen described the collaboration in an October 29 announcement. Its starting point is a capacity problem: limited electricity supplies can restrict AI expansion, raise costs and reduce operational flexibility. The companies’ proposed response is to move computing toward available power, rather than rely only on bringing more power capacity online.
Emerald AI founder and CEO Varun Sivaram described a specific intended use: easing a data center’s electricity demand during rare periods of grid stress while continuing to serve customers. That is the collaboration’s stated goal, not a result demonstrated by this proposed pilot.
The work targets AI inference—the computing used to answer requests with an AI model. Sivaram said the companies are developing routing that would direct those requests to locations with available power while keeping delays low and performance high.
The proposed integration pairs Emerald’s Conductor platform with Lumen’s programmable fiber network and edge locations. It would extend Emerald’s power-aware technology beyond decisions inside a data center and into network routing: the network would help determine where the computing happens.
Electricity availability would not be the only consideration. The companies say routing must also account for network performance, latency—the delay in serving a request—security and cost. The intended system would therefore balance access to power against the requirements of delivering the AI service.
The network has traditionally connected computing resources wherever they happen to be. We’re exploring how to turn that equation around, making the network an active part of how and where computing happens.
Jim Fowler, Lumen chief technology and product officer
For now, Lumen and Emerald are exploring a test using Lumen’s edge and network infrastructure, rather than announcing a finished routing service. The proposed pilot would test whether the combined approach can make workload placement responsive to energy conditions while accounting for the other demands on the network.
Loading discussion...
Join the conversation
Explain which uses could tolerate a delay—and which could not.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.