Thursday, July 10, 2025

Agentic AI Should Change Computing Infrastructure: Issue is How Much

Agentic artificial intelligence, eventually featuring teams of autonomous agents working in concert, should have some obvious impact on computing infrastructure. 


Chips will shift further in the direction of custom silicon. There will be more need for low-latency networking; more local or edge processing in addition to remote processing; more parallel and dynamic context to processing; distributed and fault-tolerant processing; more access to distributed databases. 


Specialized hardware (graphics processing units and field programmable gate arrays); more orchestration and more security also will be needed. Think perhaps of swarms of autonomous drones that have to work together, for example. 


In general, we will need “more:” more energy; more chips; more networking; more processing; more interworking and collaboration between autonomous systems. 


So how does that look for a firm such as Lumen Technologies, as a supplier of networking? Perhaps nobody doubts that “more” capacity will be needed, and might be needed in some different locations. 


The issue might be “how much” AI networking requirements actually change market demand, aside from the obvious “more capacity” that is continually needed. 


For starters, Lumen is doubling its intercity fiber mileage; upgrading bandwidth to 100 Gbps and 400 Gbps, using self-provisioning for enterprise customers, with plans to upgrade to 1.2 Tbps to 1.6 Tbps. 


Lumen also is building private networks that connect data centers owned by hyperscalers. But it might be the change in where capacity is needed that will change most. For some time, networking capacity has been driven both by the need to interconnect data centers and the need to make more bandwidth available in the access network so end users are connected with sufficient bandwidth and low-latency services. 


Agentic AI does not necessarily change that situation. Data center interconnection will drive developments in the network backbone. And AI used by edge devices will continue to rely on “on the device” local processing. But requirements for more edge processing in addition to “on the device” will likely mean more regional data center computing and therefore more bandwidth of a regional nature. 


Whether peer-to-peer requirements lead to more meshy architectures remains to be seen. But to some extent agentic AI simply continues other trends such as needs for more symmetrical bandwidth in the access network. As upstream bandwidth became more important as users started routinely uploading images and video, so agentic AI will additionally create more need for bidirectional capacity as local processors and actions combine with web services, software as a service platforms and application programming interfaces.


Barring a big change, such as Lumen somehow divesting its entire local telecom business, to become a latter-day Level 3 Communications capacity supplier, AI-driven requirements might be more incremental than disruptive.


As a financial matter, a Lumen that is a pure-play capacity provider might have 70 percent of present revenue, but a higher valuation. Some believe that could result in a Lumen valuation that is up to double what the firm presently commands, assuming "flawless execution" and probably also hinging on how the debt burden gets distributed.

No comments:

"Speeds and Feeds" for Home Broadband: It's the PC Story

Though definitions of “broadband” matter for regulators, advocates and suppliers, in most cases “broadband capability” matters quite little ...