AI raises heat on Hong Kong data centres

The Sandy Ridge cluster could consume 1.75 billion kWh a year.

Hong Kong data centre operators may need to go beyond mandatory energy efficiency requirements as artificial intelligence (AI) increases power consumption and heat in a city warming more than twice as fast as before, analysts said.

“The government or data centre owners may consider setting absolute energy, water, and carbon targets, in addition to those for efficiency,” Harry Lai, executive director of the Hong Kong Green Building Council, said in an emailed reply to questions.

The government will extend mandatory energy efficiency requirements to data centres from Sept. 20, requiring energy audits at least once every five years and disclosure of technical information.

Operators will not be required to implement measures identified by those audits, and the government has not set a limit on how efficiently data centres use power.

Efficiency measures alone may not show whether facilities are consuming more resources as they expand, Lai said.

'Rising heat'

Hong Kong’s rising temperatures add to the cooling challenge. The city warmed by 0.35°C per decade from 1996 to 2025, more than twice the rate recorded over the longer period from 1885 to 2025, according to the Hong Kong Observatory.

The first half was also unusually warm, with the average low of 21.5°C the highest recorded for the period and the average high of 26°C the second highest.

Traditional air cooling is becoming less suitable for AI systems, which can use up to eight times as much power per server rack as older equipment, said Chris Howard, head of account management for data centres at Jones Lang LaSalle Ltd. (JLL) in the Asia-Pacific region.

Operators are shifting to liquid cooling, which removes heat closer to processors, Howard said in an email. Backup systems are needed because temperatures can rise quickly if liquid flow stops, he pointed out.

AI systems might also require more floor space for batteries, electrical equipment, and transformers because of their higher power requirements, he said.

The concentration of several facilities in one area could add to local heat loads, said Shipeng Yan, assistant professor of management and strategy at The University of Hong Kong.

The government broke ground in March on the Sandy Ridge Data Facility Cluster in the Northern Metropolis. The cluster could consume about 1.75 billion kilowatt-hours of electricity a year, according to a March estimate by The Green Earth Ltd.

“Data centre design should also consider physical climate risks,” Yan said in an emailed reply to questions.

'Power response'

Companies are already preparing for the higher power and cooling demands.

For Sandy Ridge, CLP Power Hong Kong Ltd. (CLP Power) has formed a dedicated task force to plan and deliver the electricity infrastructure needed for the cluster, working with data centre operators on their expected power requirements.

Data centres already account for more than 7% of CLP Power’s electricity sales, with sales to the sector rising 11.8% year-on-year in the first half of the year.

Ka Wing Tang, director at Equinix Hong Kong Ltd., said its HK6 data centre adjusts cooling automatically based on temperature and computing demand.

HK6 also separates hot exhaust air from cooled air and supports direct-to-chip liquid cooling for high-density AI deployments, Tang said in an emailed reply to questions.

More efficient cooling might not offset higher electricity consumption from AI equipment, said Tsang Hon-ki, dean of the Faculty of Engineering at The Chinese University of Hong Kong.

“Other considerations could be the carbon footprint for measuring the environmental impact of data centres,” he said in an email.

Edwin Lau, founder of The Green Earth, said data centre electricity use rose 75.5% from 2018 to 2023, whilst related carbon emissions increased 35.6% to 922,000 tonnes.

Lau said operators should disclose total electricity and water use, emissions, and actual efficiency levels to show whether efficiency gains are keeping pace with expansion.

“Transparent data would also open access to green finance for retrofits and enable corporate customers to select higher-performing facilities,” he added.