Pengcheng Cloud Brain II

Background

Pengcheng Cloud Brain II is a large-scale AI computing infrastructure in China, jointly developed by Pengcheng Laboratory and Huawei.

Designed to support next-generation AI research and fundamental breakthroughs in information processing, it enables both theoretical innovation and real-world applications across multiple industries. With 1000P computing power (equivalent to 10¹⁸ operations per second), it provides a critical foundation for large-scale AI development.

As a high-performance AI data center, it requires continuous, stable operation, placing strict demands on cooling reliability and efficiency.

Key Challenges

For large-scale AI data centers, balancing reliability, efficiency, and safety is critical.

• Energy efficiency requirement: PUE must remain below 1.25
• Thermal risk under power failure: temperatures can rise 10–20℃ within one minute, risking IT equipment and system stability
• Uninterrupted operation: Cooling must run 24/7, year-round
• High-density computing: Increasing demand for advanced cooling solutions

MBT Solution

MBT delivered a highly integrated cooling solution, combining multiple systems into a coordinated and efficient platform:

• High-efficiency centrifugal chillers : 3 centrifugal chillers (2461 kW each), total capacity meeting 4428 kW demand
• Redundant system design: “2+1” redundancy across chillers, pumps, and cooling towers, ensuring high reliability, fast heat transfer, and low-noise operation
• Liquid cooling deployment: Cold plate liquid cooling system (CDU + racks), independent cooling loop
• Intelligent control system: Automatic mode switching based on outdoor wet-bulb temperature, integrated with intelligent diagnostics and centralized management: Mechanical cooling (TS ≥ 20°C), Pre-cooling (14°C < TS < 20°C, using plate heat exchangers), Free cooling (TS < 14°C, maximizing use of ambient cooling sources)
•  Emergency cooling assurance: Thermal storage tanks + UPS-backed pumps ensure short-term cooling continuity
• Modular integration: Cooling source, chilled water, and cooling water systems are integrated into modular units, enabling compact layout, faster installation, and efficient system operation

Project Value

Reliable Operation

• Redundant architecture ensures continuous system availability
• Emergency cooling reduces risks during unexpected shutdowns

Energy Efficiency

• Achieved annual PUE < 1.25
• Liquid cooling achieves approximately 30% energy savings

Intelligent Optimization

• Dynamic mode switching based on ambient conditions
• Improved system efficiency under varying loads

Delivering Reliable and Efficient Cooling for AI Data Centers

The Pengcheng Cloud Brain II project demonstrates how an integrated cooling approach can effectively address the stringent requirements of large-scale AI data centers.

By combining high-efficiency equipment, redundant design, liquid cooling technology, and intelligent control, MBT ensures stable operation while significantly improving energy performance.

This project provides a proven reference for developing reliable and energy-efficient cooling solutions in next-generation data centers.