|
Work item:
|
L.Arch_AI energy
|
|
Subject/title:
|
Specification of energy supply system architecture for Artificial Intelligence computing scenarios
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2026 (Medium priority)
|
|
Liaison:
|
-
|
|
Supporting members:
|
CAICT，China Unicom, China Tower Corporation, China Telecom
|
|
Summary:
|
The rapid advancement of large-scale AI models has sparked significant concerns over their enormous energy consumption, particularly during the training phase. As models grow in size and complexity, their computing power demands are doubling every few months, far outpacing gains in energy efficiency. Widespread use of such models in inference—answering billions of queries daily—could further escalate energy use. This rising energy footprint underscores the urgent need for more efficient algorithms, specialized hardware, new architecture of energy supply in data centres to ensure sustainable AI development.
In this regards, this Recommendation specifies energy supply system architecture for AI computing scenarios, which defines configurations of the system, basic requirements for main components and best practices. The unified standard will provide a essential blueprint for manufacturers, data centre operators, and utility providers, enabling interoperability, driving efficiencies in component design, and facilitating the seamless integration of renewable energy sources directly into high-density compute data centres.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2025-11-12 07:04:18
|
|
Last update:
2025-11-12 07:09:57
|
|
