SenseTime Lingang AIDC Wins CAICT's First AAAAA Compute-Power Coordination Certification, China's Only Dual-5A Center

CAICT announced the evaluation results at the 2026 China Computing Conference during a forum on zero-carbon foundations and compute-power coordination. CAICT Chief Engineer He Baohong presented the certificate to SenseTime. In 2026, CAICT launched a compute-power coordination testing system covering infrastructure, system platforms and scenario verification, and is developing standards for general foundations, equipment and facilities, platform scheduling and green low-carbon areas, the report said.

The test assessed economic capability, reliability capability and operation and management capability through 11 items. It covered electricity bill optimization through storage peak-valley arbitrage, park load peak shaving and valley filling, response to grid demand, storage safety protection including hardware safety, fault ride-through and emergency backup power, and platform scheduling and maintenance including data monitoring, battery state assessment, multi-system two-way connection and tiered alarm protection. SenseTime Lingang AIDC passed all items and received the top AAAAA level.

Electricity is becoming a key factor in scaling AI infrastructure. To maintain stable computing supply while improving energy efficiency, reducing power costs and strengthening grid coordination, SenseTime's AIDC worked with CATL to build a 17.888 MW/35.776 MWh new energy storage system. SenseTime also independently developed a compute-power coordination platform and deployed it first at Lingang AIDC. The platform integrates an energy model and storage system and builds an eight-level data penetration system and five intelligent decision chains, linking compute scheduling, electricity price forecasting, storage optimization, capacity control, energy management and token output optimization.

The platform prioritizes data connectivity. It connects five systems: compute scheduling, power distribution monitoring, storage BMS, HVAC cooling control and power trading interfaces. It establishes an eight-level data link from the park gateway to individual GPUs, allowing computing tasks to be traced and energy consumption to be attributed. For forecasting, the platform uses 24-hour rolling forecasts for IT load, park gateway total load and computer room PUE, and uses the results for storage charging and discharging, demand control and electricity declaration. The report said compute load forecast accuracy is stable above 96 percent.

For regulation, the platform combines storage and computing as flexible park resources. When a demand response instruction arrives, storage discharges first at millisecond level; if that is insufficient, it reduces low-priority offline computing tasks to meet the grid's downward adjustment target. In electricity price analysis and trading, the platform collects time-of-use prices, spot prices and demand response compensation prices, calculates multi-scenario returns, outputs an economically optimal storage schedule and connects to power trading declaration interfaces. It converts load forecasts into day-ahead and monthly electricity declaration references, with a declaration deviation rate within plus or minus 2 percent, the report said. It also uses TPW, or tokens per watt, as a core management metric alongside PUE, total token output, energy consumption per unit of computing power and peak-valley electricity ratio.

The platform has produced results. In energy efficiency, the park's annual PUE is expected to reach 1.24 in 2026, better than the national energy efficiency threshold of 1.25 for new large data centers, and annual carbon reduction is expected to reach 24,000 tons per 10,000P based on the current scale. On economics, coordinated optimization raised token output per unit of electricity cost by 80 percent, and the average electricity price was about 10 percent lower than that of IDCs in the same region. On grid coordination, Lingang AIDC has more than 25 MW of demand response capacity. In Shanghai's 2026 summer peak power demand response, its demand response volume and fulfillment rate both ranked first in Shanghai's data center industry and set a record, according to the report.

SenseTime said the certification verifies the technical path and industrial value of coordinating AI computing power and electricity. It plans to continue integrating computing power, electricity and AI applications and to upgrade intelligent computing centers from scale construction to lean operation, providing practices for greener, more efficient and more intelligent AI infrastructure.