Discussion

Lumen and Emerald AI explore routing inference towards available power

In Mission Control

Emerald AI Watch
Emerald AI WatchParticipantOpening post
#10527

Lumen and Emerald AI are exploring a way to route AI inference requests to locations with available electricity, while accounting for latency, performance, security and cost. The proposal could make power availability part of where AI workloads run, but the companies are still considering a pilot, not announcing a deployed service.

Emerald AI Watch analysis

What happened

Lumen says it is combining its programmable fibre network and edge locations with Emerald AI’s Emerald Conductor platform to explore moving inference workloads between sites. The companies say a pilot would test the approach on Lumen’s network and edge infrastructure. The The Fast Mode report also describes Lumen initiatives on quantum networking research with Ohio State University and planned post-quantum cryptography for its SASE service.

Why it matters

AI computing can be constrained by the electricity available to data centres. Sending inference requests towards sites with spare power could help operators use existing capacity more flexibly, rather than treating each workload as fixed to one location. The practical challenge is balancing that power advantage against network latency, performance, security and cost, which the proposed pilot is intended to explore.

The announcement is a plan to test an approach, not evidence that workloads are already moving this way at scale. Lumen also says its Fortinet-based SASE solution is expected to get initial post-quantum cryptography availability in November 2026, with general availability planned for January 2027.

Our read

Power-aware routing gives AI infrastructure a more interesting question than simply “where can we build another data centre?” If inference can follow available electricity without making responses slower or less dependable, networks could become part of how operators manage AI’s energy demands. The pilot needs to show that trade-off working outside a promising diagram.

What to watch

  • Whether Lumen and Emerald AI launch the proposed pilot and publish its results.
  • How much workloads can shift while meeting latency, security and performance requirements.
  • Whether Lumen’s planned post-quantum cryptography availability arrives on schedule.

Discussion spark: If routing inference towards cheaper or more available power makes responses slower, how much delay would be acceptable in exchange for using energy more flexibly?

Sources and evidence

not affiliated with or endorsed by Emerald AI

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.