China’s First AI Computing–Power Coordination White Paper: DeepCtrls Advances Infrastructure with Physical AI
Published: 2026-9-2
On September 2, the Research Report on Artificial Intelligence Applications for Green Computing was officially released at the opening ceremony of the 2026 Open Data Center Conference and inaugural ODX Expo. The report was jointly led by DeepCtrls and the Cloud Computing and Big Data Research Institute of the China Academy of Information and Communications Technology (CAICT), with contributions from industry partners including CATL, Foxconn Industrial Internet, Shanghai Unicom, and Chindata.
As China’s first white paper focused on AI-enabled computing–power coordination, the report draws on the contributors’ extensive industry experience and technical expertise. It establishes a systematic technology framework and application roadmap for AI-powered computing infrastructure, providing methodological guidance for the intelligent evolution of next-generation facilities.
With Physical AI at its core, the report covers the full infrastructure lifecycle, including planning and design, operational optimization, energy coordination, and maintenance management. It addresses key areas such as AI-driven energy optimization, dynamic computing–power coordination, intelligent maintenance, and liquid cooling control. Together, these approaches advance green computing from individual energy-saving measures to system-wide optimization, from static management to dynamic adjustment, and from experience-based maintenance to intelligent decision-making.
As AI clusters with tens of thousands of accelerators are deployed at increasing speed, computing infrastructure is becoming denser, more complex, and more dynamic. Tight coupling between computing, power supply and distribution, cooling, and energy storage makes conventional management based on operator experience and fixed strategies increasingly inadequate.
The future of green computing lies in infrastructure that can sense, predict, make decisions, and optimize itself—progressing toward intelligent autonomy.
Drawing on years of practical deployment, Deepctrls has developed a technology approach that uses Physical AI to transform computing and energy infrastructure. It combines physics-based frameworks and constraints with machine learning and reinforcement learning, enabling AI to understand equipment and system behavior and continuously identify optimal operating strategies within safety limits.
In energy optimization, Physical AI monitors system conditions in real time, combines load forecasts with model-based analysis to generate optimization strategies, and directs control systems to execute them.
For computing–power coordination, Deepctrls connects energy-side control interfaces—including BA, PLC, and PCS—with computing orchestration and scheduling interfaces such as Kubernetes (K8s) and Slurm. This enables coordinated decisions and closed-loop execution across computing, power, cooling, and energy storage. The result is a complete operational chain: AI generates optimized solutions, physical constraints maintain safety, and control systems execute the decisions.
The report presents six categories of real-world engineering applications from Deepctrls’ work in AI and conventional data centers: coordinated supercomputing optimization, PUE diagnostics for water-cooled plants, demand-based air cooling, indirect evaporative cooling maintenance, coordinated air and liquid cooling, and simulation-based design of high-efficiency cooling plants.
In a combined air and liquid cooling data center project, Deepctrls used Physical AI to coordinate cooling sources, coolant distribution units (CDUs), racks, and air-cooling systems. Cooling capacity was dynamically matched to actual heat dissipation requirements, achieving average electricity savings of 10.3% in short-duration comparative tests. An indirect evaporative cooling project is expected to deliver 13.6% electricity savings for air handling units (AHUs).
The report represents a joint industry effort to advance green computing through Physical AI and consolidates Deepctrls’ years of practical experience. It also marks the company’s progression from technology development and product deployment to contributing industry-wide methodologies.
Deepctrls will continue integrating Physical AI into computing and energy infrastructure, building a general-purpose AI brain for infrastructure in the AI era and advancing green computing toward a new stage of intelligent operation powered by Physical AI.
