AIDC Infrastructure Specification Interpretation Report

This page provides a structured interpretation of the AIDC Infrastructure Specification v0.8, covering standard positioning, applicable scope, power classification, framework, power distribution and liquid-cooling requirements, energy-efficiency targets, implementation guidance, and industry value.

Official Version V0.8

Single-Rack Target 300kW

Efficiency Target PUE ≤ 1.15

Publishing Body GCC

1. Standard Overview

The AIDC Infrastructure Specification, short for Artificial Intelligence Data Center Infrastructure Specification, is the industry's first systematic technical standard focused on the construction of artificial intelligence data centers. The specification was pre-released on September 17, 2025 at the first AIDC Industry Development Conference and officially released as version 0.8 in January 2026 after approval by the technical steering committee on January 9, 2026.

1.1 Positioning and Background

As large-model parameter counts continue to grow explosively, multimodal applications keep expanding, and intelligent inference demand rises sharply, computing scale is increasing at an exponential pace. At the same time, interconnect scale and equipment density are also rising rapidly. This unprecedented wave of computing demand creates disruptive challenges for AI data center infrastructure. The traditional progressive design model used for conventional data centers, which assumes relatively stable loads, is facing a systemic adaptation gap under the high-volatility, high-density, and fast-iteration characteristics of AI workloads.

1.2 Standard Information

Item	Content
Standard Name	AIDC Infrastructure Specification
Version	V0.8 (Official Release)
Publishing Body	Global Computing Consortium (GCC) - Open AI Infra Community
Release Timeline	Pre-release: September 17, 2025; Official release: January 2026
Lead Organization	Yiqi Research Institute
Standard Type	Association Standard / Community Specification

1.3 Participating Organizations

The specification was led by Yiqi Research Institute and jointly compiled by dozens of leading companies across internet platforms, telecom operators, server manufacturing, chips, design institutes, and IDC infrastructure. Core contributors include Huawei, ByteDance, Kuaishou, JD.com, Baidu, xFusion, China Comservice, CEI, GDS, VNET, Sugon, Lenovo, China Mobile, Runze Technology, Envision Energy, Xiaohongshu, Bilibili, H3C, Envicool, Vertiv, GLP, Haoyang Cloud Data, Schneider Electric, and Sanhe Tongfei.

2. Scope and Core Objectives

The specification is designed for full-scope artificial intelligence data center scenarios. It takes an integrated design approach spanning campuses, buildings, and floors, coordinating building structure, power distribution, heat dissipation, and network cabling within one framework.

2.1

Core Objectives

Meet the expected single-rack power evolution target of about 300kW within the next three to five years.
Enable decoupled design between the data hall and the rack.
Support the evolution of two to three generations of AI liquid-cooled integrated racks.

2.2 Power Classification

Level	Single-Rack Power Range	Typical Application Scenario
Level 1	100kW	General AI inference and lightweight training
Level 2	200kW	Medium-scale training and mixed workloads
Level 3	300kW	Ultra-large-scale training and high-performance computing

Design Principle

Air cooling, liquid cooling, and power distribution are all defined through graded interfaces. This allows infrastructure to flexibly support products from 100kW to 300kW and, at its core, builds a long-term interface system that decouples the room infrastructure from rack equipment.

3. Framework of the Specification

The AIDC Infrastructure Specification covers the full life cycle of AIDC infrastructure delivery and forms an eight-chapter structure from general rules to operations and maintenance.

3.1 Framework Overview

Chapter 1 General Rules: Purpose, basis, scope, terminology, and definitions.
Chapter 2 General Requirements: Design principles, performance baselines, and interface specifications.
Chapter 3 Room Environment Requirements: Temperature, humidity, cleanliness, vibration control, and electromagnetic environment.
Chapter 4 Building Requirements: Building structure, floor loading, floor height, and logistics channels.
Chapter 5 Power Distribution System: Power architecture, distribution design, UPS, and backup power.
Chapter 6 Cooling System: Liquid cooling, cooling water system, air-conditioning and airflow, and efficiency management.
Chapter 7 Network Cabling: Optical backbone, copper cabling, rack wiring, and interface standardization.
Chapter 8 Delivery and O&M: Delivery process, acceptance criteria, operations management, and emergency planning.

4. Core Technical Requirements

AIDC infrastructure sets a higher bar for power distribution, cooling, civil works, and network design. Under high-density AI deployment scenarios, liquid cooling is no longer just an auxiliary option and has become a key configuration direction.

4.1

Power Distribution Requirements

High reliability, distributed architecture, and standardized interfaces are the core principles.

Incoming voltage: 10kV medium-voltage input.
Distribution voltage: 800V high-voltage DC optional.
Architecture: 2N redundancy.
Power reliability: ≥99.99%.
Planning is based on a full 300kW rack configuration to support long-term evolution.

4.2

Cooling System Requirements

Liquid cooling is the key differentiator between AIDC and conventional data centers.

Flow-control accuracy: ±5%.
Pressure loss: ≤0.1MPa.
Heat-dissipation redundancy factor: ≥1.2.
Coolant temperature control: 40±2℃.
GPU core temperature: 65℃±2℃ for cold-plate cooling.

4.3

Building and Structural Requirements

Civil planning must target the era of high-density liquid-cooled racks from the very beginning.

Floor loading is designed for a full 300kW rack configuration.
Floor height must accommodate liquid-cooling pipes, busways, and integrated cabling.
Logistics channels must support large equipment movement and rapid delivery.
Modular design is encouraged to avoid large-scale facility rebuilds when equipment changes.

4.4

Network Cabling Requirements

The design must support low-latency and high-bandwidth interconnection for AI clusters.

Compute network: ≥40Gbps.
Storage network: ≥100Gbps.
Management network: ≥10Gbps.
Optical backbone, standardized interfaces, and modular deployment improve compatibility and installation efficiency.

Cold-Plate Liquid Cooling

This route is suitable for existing room retrofits and medium-density scenarios. It is easier to deploy and can extend GPU life by more than three times.

Immersion Liquid Cooling

This route is suitable for new supercomputing centers and high-density deployment scenarios, with PUE potentially as low as 1.07.

Comparison with Traditional Air Cooling

Traditional air cooling usually supports about 20kW per rack, while AIDC liquid-cooling solutions can support 300kW or above, with clear improvements in cooling efficiency and energy-use share.

5. Green and Energy-Efficiency Requirements

The specification sets PUE≤1.15 as a mandatory access threshold and also brings carbon-emission control, waste-heat utilization, and independent verification into one unified framework.

5.1 PUE Requirements

Policy / Region	PUE Requirement	Notes
National Green and Low-Carbon Development Action Plan for Data Centers	≤1.25 by the end of 2025	Large data centers
AIDC Specification	≤1.15	AI computing data centers
Beijing / Shanghai tiered electricity pricing	Surcharged when exceeded	Drives technical upgrades

Energy consumption of substations, cooling towers, and other supporting facilities is fully included.
Compliance loopholes caused by an artificially enlarged denominator are prevented.
An independent verification mechanism is required.

5.2 Carbon Emission Control

Indicator	Requirement
Renewable energy utilization rate	≥30%
Energy-storage system response delay	≤50ms
AI training carbon-intensity	≤0.2kgCO₂/hour

Case: after adopting the new AIDC standard, a pilot intelligent computing center in Shanghai reduced energy consumption by 40% for every 1PFlops of computing handled, saved 12 million kWh annually, and reduced carbon emissions by about 6,800 tons.

1.15 PUE access threshold in the AIDC specification

30% Required renewable-energy utilization rate

50ms Energy-storage response target

6. Relationship with Other Standards

The AIDC Infrastructure Specification forms a complementary system with conventional data-center standards and shows clear differences in rack power, liquid-cooling configuration, build cycle, and design philosophy.

6.1 Relevant National Standards

Standard No.	Standard Name	Relevance
GB/T 32910.3-2016	Data Center Resource Utilization Part 3	Basic standard for data-center energy efficiency
GB/T 22239-2019	Information Security Technology Baseline for Classified Protection	Basic security compliance requirement
GB 50174-2017	Code for Design of Data Centers	Conventional data-center design standard
YD/T 3005-2016	Design Specification for Data Center Construction (Telecom Industry)	Telecom-sector data-center standard

6.2 Differences from Conventional Data-Center Standards

Dimension	Conventional Data Center	AIDC Infrastructure Specification
Single-rack power	3-10kW	50-300kW
Cooling mode	Mainly air cooling	Liquid cooling required
PUE target	1.8-3.0 / 1.2-1.4	≤1.15
Construction cycle	6-8 months	3 months with standardized design
Design philosophy	Progressive and locally optimized	Globally restructured and decoupled
Applicable load	Stable workloads	High-volatility, high-density, fast-iteration workloads

7. Implementation Points and Frequently Asked Questions

The specification applies to intelligent computing center developers, conventional data-center retrofit projects, equipment suppliers, design and construction teams, and O&M service providers. The key is to combine full-capacity planning at the front end with ongoing optimization during operation.

7.1 Planning and Design Stage

Demand analysis: define computing scale and power-density level.
Site assessment: evaluate energy supply, network conditions, and climate.
Architecture design: adopt modular and standardized design concepts.
Elastic planning: reserve expansion space for future two-to-three-generation evolution.

7.2 Construction and Delivery Stage

Civil preparation: configure floor loading and floor height for full-capacity power targets.
System integration: coordinate liquid cooling, power, and network deployment.
Standardized delivery: follow the interface standards defined by the specification.
Acceptance testing: verify PUE, energy efficiency, reliability, and other core indicators.

7.3 Operation and Maintenance Stage

Energy-efficiency monitoring: track PUE and carbon emissions in real time.
Intelligent scheduling: dynamically optimize resource allocation.
Predictive maintenance: prevent failures through data-driven analysis.
Continuous optimization: iterate using operational data.

Frequently Asked Questions

Q1: What is the core difference between AIDC and a conventional data center? The core difference lies in the design philosophy. AIDC takes decoupling between the data hall and the rack as a central concept and reconstructs infrastructure around hard constraints such as mandatory liquid cooling and PUE≤1.15.
Q2: How can the initial investment in liquid-cooling systems be controlled? It can be optimized through cold-plate liquid cooling for retrofit scenarios, modular phased deployment, and long-term efficiency gains that offset the initial investment.
Q3: How can compatibility with future equipment be ensured? By designing civil conditions to a full 300kW target and using standardized interfaces that support the evolution of two to three generations of AI liquid-cooled integrated racks.
Q4: How can PUE≤1.15 be achieved? It requires coordinated measures such as liquid cooling, high-voltage DC distribution, intelligent thermal control, and waste-heat recovery.

8. Standard Value and Appendix Information

The significance of the AIDC Infrastructure Specification lies in driving the AIDC industry from isolated technical breakthroughs toward standardized and system-level development while continuously targeting the next two to four years of technical evolution.

8.1 Three Core Values

It addresses the bottleneck of waiting for machine-room readiness and shortens facility completion cycles from 6-8 months to about 3 months.
It lowers total industry cost by promoting large-scale adoption of efficient technologies such as liquid cooling and supports low-carbon development.
It guides orderly competition by building consensus among leading companies and defining clear interface standards and collaboration frameworks.

8.2 Compilation Features

Full-stack ecosystem participation across the industrial chain.
Deep user involvement, with demand-side pain points driving collaboration.
Open co-creation, welcoming continued community contribution.
Dynamic updating, with annual revision plans targeting future two-to-four-year trends.

Reference Links

GCC-Open AI Infra Community: https://globalcomputingconsortium.org
Specification download: visit the GCC official website or obtain it through the official QR code.

Glossary

Term	English	Definition
AIDC	Artificial Intelligence Data Center	An artificial intelligence data center.
PUE	Power Usage Effectiveness	A metric for data-center power efficiency.
GCC	Global Computing Consortium	The Global Computing Consortium.
Cold-plate liquid cooling	Cold Plate Liquid Cooling	A liquid-cooling method that removes heat through cold plates in contact with hot components.
Immersion liquid cooling	Immersion Liquid Cooling	A cooling method in which equipment is fully immersed in coolant.
Decoupled design	Decoupling Design	A design concept that separates infrastructure capability from equipment evolution.

Note

This report is compiled from public information for reference and learning purposes. For the original specification, please visit the official website of the Global Computing Consortium.

Greentest Service Account
Greentest Subscription Account

Tel: 020-2204 2442

E-mail:Sales@greentest.com.cn