(Yicai) Aug. 17 -- Ulanqab in Inner Mongolia Autonomous Region has become one of the most robust artificial intelligence data center clusters in China, with the city's low electricity costs amid abundant renewable energy resources and cooler climate attracting industry giants like Alibaba Group Holding.

"For data centers of the same scale, the annual electricity costs in Ulanqab can save CNY5 billion (USD741.5 million) compared to operating in other cities," Wang Zhaoyang, general manager of Alibaba Cloud's local AI data center, told Yicai. Zhangjiakou in northwestern Hebei province was home to the firm's largest base in the cloud computing era, but Ulanqab has far outperformed others in the AI computing era, Wang added.

Visitors at Ulanqab's Chahar High-Tech Zone can spot more than 30 data centers in just a 10-minute drive, Yicai noticed. In addition, construction workers can be seen across many data center parks, as new facilities keep rapidly appearing.

Market proximity, cost advantages, and technological demands were the three interrelated factors that made Alibaba Cloud pick Ulanqab as its primary base of operation, Wang noted. Data centers need to be close to end markets, especially for latency-sensitive applications such as quantitative trading and online games, Wang pointed out, but stressed that the low electricity prices are the most decisive factor.

Electricity prices in Ulanqab range from 32 Chinese cents to 35 Chinese cents (around 5 US cents) per kilowatt-hour, while the city's average annual temperature of 4.3 degrees Celsius is conducive to heat dissipation. It also sits only about 350 kilometers away from Beijing, where many Chinese users are based, so the round-trip network latency is merely 4 milliseconds.

Yicai also visited Alibaba Cloud's innovative data center designed exclusively for the AI era. The facility was built based on the company's fully modular CUBE 5.0 architecture, which reduces construction time for AI data centers to just 100 days from the standard six to 12 months.

"Faster delivery equals faster productivity," Wang noted. "No computing card sits idle, as each is either training AI models or generating tokens, with early delivery providing extra value to clients."

Editors: Tang Shihua, Martin Kadiev