Hong Kong Data Centre Prices Jumped 90% This Year. The Constraint Isn’t Demand.

Hong Kong's data centre market is producing one of the sharpest price movements in Asia, and it is doing so for an unflattering reason: the city cannot deliver capacity at the speed demand is arriving.

Wholesale pricing bands offered to hyperscale operators have risen 90% since the beginning of the year, among the largest increases anywhere in the region. Rates now reach as much as $180 per kilowatt — still well below the $300 to $490 range in Singapore, where supply constraints are more severe, according to Structure Research.

That gap is worth holding onto. Hong Kong is expensive and getting rapidly more so, but it has not yet reached the level at which Singapore's constraints priced that market out of contention.

Why Chinese AI firms want Hong Kong specifically

The demand driving this has a particular origin, and it is not primarily international hyperscalers.

Hong Kong offers Chinese technology firms something the mainland cannot: free flow of data and direct access to international markets, in contrast to China's tightly controlled internet environment. One analyst characterised the city as a useful cocoon in which to train AI systems and test them against international competition.

For a Chinese AI company with global ambitions, that combination — mainland proximity, international connectivity, no Great Firewall — has no substitute in the region.

Operators are financing accordingly. SUNeVision Holdings, DayOne Data Centers and Equinix are among those raising billions of dollars in debt to fund expansion.

The supply side is the bottleneck

The physical position explains why prices are moving rather than volumes.

Hong Kong operates approximately 581 megawatts of live data centre capacity, with more than 400 MW under construction or in planning, according to figures cited by the British Chamber of Commerce in Hong Kong. Around 54 colocation facilities are operational, most built to Tier III standards, with Tseung Kwan O the dominant cluster — accounting for roughly 41% of existing white floor space and about 26% of upcoming supply.

CBRE's assessment is that Hong Kong recorded modest growth in 2025 because of development constraints rather than any shortage of demand, even as Asia Pacific data centre investment hit a record US$11.6 billion. Growth across the region has been shifting toward power-advantaged markets including Malaysia, Australia and India.

The constraints are structural. Land scarcity forces vertical construction, which raises power usage effectiveness — more energy consumed on cooling and non-computing functions than a horizontal campus would require. Ensuring reliable power demands substantial investment in distribution systems and backup generation, and sourcing renewable supply adds further complexity.

Hong Kong does have genuine advantages: a dual-grid power supply with high reliability, low seismic and typhoon risk relative to some regional peers, and adequate fresh water permitting water cooling.

Government is moving on land

The policy response has been to release sites, and the largest is substantial.

The government awarded the Sandy Ridge Data Facility Cluster site on 2 March 2026 for development into an advanced data facility cluster. Reporting on the award describes 11 hectares granted to a Chinese computing technology company, with cumulative investment estimated at HK$23.8 billion and expected capacity of 180,000 PFLOPS by 2032 — around 36 times Hong Kong's current computing power.

Private investment is arriving alongside it. Equinix opened HK6, its sixth Hong Kong facility, in June with an initial investment of US$124 million, providing 1,000 cabinets in the first phase and scaling to 3,550 at full build-out. The facility uses liquid cooling, which AI workloads increasingly require rather than prefer, and is positioned around direct connectivity into the Hong Kong-Shenzhen innovation corridor.

Connectivity is being upgraded in parallel. HKT is building data centre interconnection on hollow core fibre between Lok Ma Chau Loop and Tseung Kwan O, targeting roughly 30% lower latency than conventional fibre — relevant because keeping thousands of GPUs synchronised across facilities is a latency problem before it is a bandwidth problem.

Inference, not training

The most useful framing of Hong Kong's position may be that it is not competing for the same workloads as its neighbours.

Structure Research argues the city is structurally better suited to AI inference than AI training, supported by its connectivity ecosystem and low-latency regional links. Training runs are power-hungry, time-flexible and can sit anywhere with cheap electricity — which is why they migrate to Malaysia and India. Inference is latency-sensitive and needs to sit near users and networks.

On that reading, Hong Kong's power constraint is less disqualifying than it first appears, provided the city targets the workloads its geography actually suits.

The demand underneath all of it continues to scale, visible in the finances of the firms generating it — OpenAI's annualised revenue has roughly doubled to top $40 billion, with further AI companies moving toward public markets.

What to watch

Whether the 90% price increase holds is the near-term question. Rising rates with flat volumes signal scarcity; rising rates with rising volumes signal a market working.

The second is Sandy Ridge delivery. A 2032 capacity horizon is a long way from a 2026 shortage, and the intervening years are when the pricing is decided.

The third is power. Every constraint described here resolves into electricity, and Hong Kong's ability to supply high-density AI facilities without compromising grid reliability is the variable that determines how much of this demand it can actually capture.