With the rise of AI computing power, data centers are undergoing fundamental redefinition.

As computing power demand skyrockets, data center infrastructure faces unprecedented challenges, making the construction of next-generation AIDC (AI Data Centers) a core proposition that the entire industry must collectively address.

On September 17, during the inaugural AIDC Industrial Development Conference, stakeholders across the industry chain—including research institutions, data center service providers, and internet operators—gathered to advance AIDC implementation and large-scale development through collaborative standards development, joint technology research, and ecosystem coordination. The conference also pre-released the "AIDC Infrastructure Specifications," providing guidance for AIDC construction over the next 2-3 years.

Miao Fuyou, CTO of the Global Computing Consortium (GCC) Secretariat, noted that as large models enter practical application phases, artificial intelligence is generating numerous new application scenarios. The AI industry is experiencing rapid application expansion, with corresponding infrastructure development in a similar phase, driving swift growth in AIDC demand.

He further indicated that based on various forecasts, domestic AIDC construction will maintain annual growth rates exceeding 40% over the next two to three years. Subsequently, annual new construction volumes will continue increasing before gradually stabilizing, with growth rates expected to reach approximately 10% by around 2030.

**Triple Challenge of Heat Dissipation, Power Supply, and Space**

However, amid rapidly rising AI computing power demands, AIDC construction also faces significant challenges.

Professor Jin Hai, Chairman of the Global Computing Consortium (GCC), highlighted that AIDC construction confronts challenges related to "heat, power, and space," along with the "AI waiting for data centers" dilemma caused by standards gaps and lengthy construction cycles. He called for accelerated industry chain standard system development, technological breakthrough achievements, and collaborative innovation ecosystem construction.

More critically, AIDC infrastructure standards remain incomplete, with upstream and downstream enterprises largely in individual exploration phases, creating cost waste and extending delivery cycles. Industry feedback commonly indicates that data center construction delivery times are often calculated in years, resulting in AI equipment waiting for facilities and impacting industry development speed.

Wang Zhiqiang, Deputy General Manager of the Data Center Business Unit at China Electronics Engineering Design Institute, explained: "The transition from cloud to intelligent computing, with data centers fully embracing artificial intelligence at the facility level, presents tremendous challenges and changes. With AI's arrival, everything from server power density and business scenarios to networking, power distribution, and cooling has changed completely from the cloud architecture era."

He revealed that the industry is accelerating the improvement of design specifications and standards in response to these changes: "New specifications and standards are being urgently revised under the AI wave's influence. They have entered the comment solicitation and review phases, with new national standards expected for release next year."

Among demand changes, power density increases are particularly noteworthy. Wang Zhiqiang noted that single cabinet power growth has far exceeded traditional design capacity: "Growth from dozens of kilowatts to 40-50KW has become standard. Huawei's recently released 384 super-node approaches 60KW per cabinet. By year-end and early next year, this capacity will reach hundreds of kilowatts, potentially reaching 200KW quickly, or even 500KW to 1000KW per cabinet—unimaginable in previous designs, planning, and overall infrastructure configurations." Therefore, new planning and design must closely follow demand.

Additionally, full lifecycle considerations pose investment and operational challenges. Wang Zhiqiang explained that servers typically iterate every three to five years following Moore's Law, while mechanical and electrical equipment lifespans range from 10-15 years, and the buildings housing this equipment may last 50-70 years: "How can we better match buildings and mechanical equipment with IT? When planning for users, especially major companies, their goal is flexibility."

Furthermore, cooling has become an unavoidable challenge in data center investment planning. Regarding cooling technology, air cooling solutions are approaching limits, making liquid cooling the transformation direction. These issues require early intervention from standards, testing, and certification systems. Planning around cluster construction challenges traditional data center design concepts.

**Standardization Acceleration Opens New AIDC Era**

Against the backdrop of coexisting challenges and opportunities, the AIDC industry is rapidly moving toward standardization and systematization. Domestic industry chains are forming alliances to explore industrial development paths starting with new standards. The conference's pre-release of "AIDC Infrastructure Specifications" represents just the beginning, with multiple key interface and component standards to follow.

Miao Fuyou emphasized that joint development of standard and specification systems is more crucial. He introduced that the alliance currently has professional committees working on liquid cooling standards, including integrated cabinet standards, interface standards between integrated cabinets and servers, and standards for key liquid cooling components such as CDU (Coolant Distribution Units).

"We will continue expanding the scope of standards, considering not only liquid cooling but also power supply standards and facility standards, gradually developing comprehensive standard systems," Miao Fuyou stated.

From an energy and power perspective, Wang Zhiqiang mentioned that data center scales and intelligent computing cluster scales are rapidly expanding from thousands to tens of thousands of cards, soon reaching 100,000-card clusters, with even larger future scales anticipated. Such massive clusters correspond to enormous electrical energy consumption, making energy conservation, consumption reduction, and green low-carbon operations critical work and the development logic pursued by industry standardization construction, core manufacturer technology roadmaps, and application scenario ecosystems.

Policy-wise, in May, the National Development and Reform Commission and National Energy Administration jointly released the "Notice on Orderly Promoting Green Power Direct Supply," providing policy support for green power direct supply and source-grid-load-storage solutions. New national standards will further strengthen constraints on PUE (Power Usage Effectiveness) and WUE (Water Usage Effectiveness) and carbon consumption, promoting standardization of liquid cooling, energy storage, and resource recovery technologies.

Another important trend in technological evolution is modular construction modes and rapid delivery. One conference expert predicted that with continued high-power equipment demand growth, "liquid cooling technology will become an ideal choice for solving high-density AI equipment heat dissipation challenges." Through optimized design processes and modular construction, the cycle from facility planning to deployment readiness could be shortened by approximately three months. This means future AIDC construction will increasingly rely on prefabricated solutions for replicable rapid deployment.

Yang Chaobin, Huawei Director and ICT BG CEO, pointed out that AIDC represents critical infrastructure for the intelligent computing era. With rapid increases in AI computing scale and chip power, liquid-cooled data centers are becoming inevitable choices for AIDC. Facing current liquid cooling facility construction challenges requires industry chain collaboration to accelerate liquid cooling facility productization and standardization, expedite AIDC deployment and utilization, and jointly promote AI industry development.

Industry observers believe that future development will utilize closed-loop systems of standard specifications, testing certification, and industrial promotion to advance AIDC from an "exploration phase" toward "standardized, replicable" new stages. For enterprises hoping to gain advantages in the next round of computing power competition, those who can first master high-density, low-energy consumption, scalable data center construction capabilities may secure first-mover advantages in the artificial intelligence wave.