Hong Kong’s unique strengths as a world financial and regional connectivity hub are helping to secure the city’s edge in the computing infrastructure sector, with the government firmly committed to making it an artificial-intelligence-geared compute base.  Chai Hua reports from Hong Kong.

In the subzero-chill city of Ulaanqab in North China’s Inner Mongolia autonomous region, abundant wind and solar power fuel one of the country’s major intelligent computing hubs. Some 3,000 kilometers away, in land-scarce, power-hungry Hong Kong, a parallel compute-capacity race is quietly unfolding — one that seemingly defies economic logic.

Hong Kong’s installed compute capacity has surged from roughly 60 peta floating-point operations, or petaflops, per second in 2023 to 5,000 petaflops this year, with ambitious plans to scale up to 180,000 petaflops by 2032 — a scale matching that of Ulaanqab.

Skepticism naturally abounds. By pouring billions of dollars into data infrastructure in one of Asia’s most expensive cities, is Hong Kong falling into an oversupply or overcapacity trap? Industry gurus tend to strike a balance.

READ MORE: HK fast-tracks Sandy Ridge AI computing hub to mid-2027

“Certain regions have attracted excessive investment and experienced temporary oversupply,” says technology strategist Charlie Dai, vice-president and principal analyst at market research firm Forrester.

He says he has yet to see a severe compute oversupply in Hong Kong as the demand from financial institutions, multinational companies, cloud providers and artificial intelligence enterprises stays healthy, although some newer facilities may experience slower utilization during their ramp-up phase.

“Premium facilities with strong power, connectivity and graphics processing unit (GPU) capabilities continue to draw customers,” says Dai. For Hong Kong, the primary risk is demand-supply misalignment rather than structural oversupply.

Mao Zhenhua, a professor of practice in economics at the University of Hong Kong Business School, says the industry’s structural mismatch of computing infrastructure usually spans three distinct dimensions — functional imbalance, geographic distribution, and market participants.

Mao, who is credited with helping to transform China into an economic superpower, explains that functional imbalance refers to an oversupply of mid- to low-end computing capacity, contrasting sharply with a shortage of high-end clusters required for large-model training.

The second factor is that AI enterprises and compute demand are sometimes concentrated in different areas. The third is market mismatch, with technology giants locked in most high-end compute resources, leaving small and medium-sized enterprises struggling with high costs, limited access, and deployment barriers.

Building AI-ready capacity

These structural challenges resonate in Hong Kong, where its traditional infrastructure is experiencing a rapid, AI-ready evolution.

Telecommunications behemoth China Mobile, which set up its global network center in the city’s Tseung Kwan O district in 2014, launched its global intelligence center in March — one of the special administrative region’s largest stand-alone intelligent computing hubs.

While the company’s global network facility leverages submarine cable assets to deliver ultralow-latency global connectivity alongside legacy internet data-center hosting and cloud services, its global intelligence center is an AI-ready computing hub featuring liquid-cooling capabilities with power densities reaching 140 kilowatts per rack to support high-throughput AI workloads. In contrast, traditional facilities operate below 30 kW per rack for general cloud requirements.

“Since the global intelligence center went into operation, the market’s response has been positive, with client onboarding and demand pipelines aligning well with expectations,” says Xiao Kang, deputy head of China Mobile International’s Enterprise Business Division.

To meet growing demand, the company’s Phase II expansion is underway, with total compute capacity expected to more than quadruple by July. However, Xiao says both centers maintain a complementary, mutually reinforcing partnership rather than a zero-sum trade-off.

Multinational digital infrastructure firm Equinix is also ramping up its investment in Hong Kong’s computing infrastructure. In June, the company launched its sixth data center in the city, HK6. With an initial investment of HK$1 billion ($124 million), the project represents Equinix’s largest single investment in Hong Kong in the past decade.

Compared to previous data centers, HK6 adds AI-ready capacity with liquid-cooling capability and high-density support to provide next-generation AI and digital services.

“AI-ready infrastructure is an evolution of the data center industry instead of being a replacement for carrier-neutral data centers,” says Joanne Hon, managing director of Equinix, Hong Kong.

She says she doesn’t believe every future data center will become a dedicated computing factory, although more facilities will need to be AI-ready, meaning greater flexibility in power and cooling design, stronger ecosystem connectivity and being able to support a broad mix of traditional and next-generation workloads.

According to Equinix, market feedback has been positive since the launch of HK6, with a “meaningful part of demand” coming from financial services and financial technology customers.

“Equinix Internet Exchange has grown by 50 percent in Hong Kong in the past year, reflecting a robust digital landscape that continues to expand,” says Hon.

As to whether recent data center developments have created a structural computing overcapacity in Hong Kong, she’s upbeat, with the market having experienced a temporary supply expansion rather than a long-term demand mismatch.

“AI is changing what customers require,” as organizations look to higher-density deployments, more advanced cooling and predictable throughput for moving large data sets securely between clouds, networks and enterprise environments, Hon says.

Hong Kong’s strategic niche

In her view, Hong Kong remains one of Equinix’s most established and sought-after markets, underpinned by its role as a global financial pivot, a regional connectivity hub, and a gateway to the Guangdong-Hong Kong-Macao Greater Bay Area.

“Hong Kong’s competitive advantage will be its ability to provide the infrastructure for local computing and AI workloads, while serving as the interconnection point that enables secure, low-latency digital exchange across the Greater Bay Area, the rest of China and the world,” she says.

Such unique dual strength is key to securing the city’s irreplaceable edge amid fierce compute infrastructure competition. While hubs on the Chinese mainland leverage green power and space for energy-intensive offline model training, Hong Kong can harness its ultralow latency, international data compliance, and neutral interconnection ecosystem to power high-value real-time AI inference, enterprise private deployments, and global corporate expansion, experts say.

Xiao says places like Ulaanqab or the cities of Shaoguan and Shantou in Guangdong province can rely on cool climates and abundant green power, making them natural hubs for energy-intensive, large-scale offline model training. In comparison, Hong Kong, with its distinct location and ecosystem advantages, is far better suited for high-value real-time inference, cross-border orchestration, application delivery and global service provision.

This represents the core strategic approach for China Mobile International. Operating within strict data compliance boundaries, the telecom giant’s real-time trading and AI inference remain localized in Hong Kong, while nonsensitive tasks like offline training and rendering are dynamically routed to cost-optimized mainland pools.

For international expansion, China Mobile  International capitalizes on high-speed submarine cables and points of presence connecting Hong Kong with Southeast Asia. Targeting regional digital growth hubs like Singapore, Malaysia, Thailand and Indonesia, the company aims to be a launchpad for Chinese AI enterprises expanding globally.

Accessibility and policy backing

Leveraging more affordable mainland computing power is also a key strategy for enterprises to control costs. After all, while embracing AI technology is a shared direction, the hefty expenses involved remain an unavoidable factor for businesses, especially SMEs.

Some service providers have launched measures to lower entry barriers for SMEs, shifting rigid leases to flexible, lightweight access, such as hourly packages, task-based bundles, and pay-as-you-go API, or application programming interface, and token pricing.

In procuring computing services, Dai suggests that enterprises should start with business requirements rather than GPU specifications. The key evaluation factors include compliance, sovereignty, performance, scalability, connectivity, service reliability, pricing transparency and security.

“The best provider isn’t necessarily that with the largest GPU resources, but one that can deliver the required business outcomes at the lowest operational risk and total costs,” he says.

Jensen Huang, who co-founded semiconductor giant Nvidia Corp and serves as its CEO, envisions that the future compute market will feature a stratified pricing ladder from a slow-response “free tier” up to high-speed and premium supercomputing tiers priced from $45 to $150 per million tokens, a price gap of hundreds of times between different tiers.

To help local businesses navigate cost tiers and prevent smaller players from being priced out, public policy is stepping in with targeted fiscal backing.

Hong Kong’s 2026 Policy Address has galvanized support for SMEs. The government’s Dedicated Fund on Branding, Upgrading and Domestic Sales will offer more targeted funding support for AI projects, while the authorities will enhance the Digital Transformation Support Pilot Programme by providing matching subsidies to assist a broader range of SMEs in adopting AI and cybersecurity solutions.

ALSO READ: Network of power

As outlined in the city’s first Five-Year Plan for Economic and Social Development (2026-30), Hong Kong is going all out to upgrade its computing infrastructure. This year’s Policy Address also said the government will commission a new consultancy study on computing power to analyze future demand.

The moves come at a pivotal moment. Hong Kong is at a critical juncture in its AI infrastructure transition — evolving from early-stage data infrastructure toward AI-ready and business-aligned computing.

Compute infrastructure is far more than a simple piling of silicon and server racks. It represents a deep, structural realignment of the industrial ecosystem.

As the SAR refines its road map, tailoring infrastructure to its strengths in finance, trade and technology can directly address concerns about structural overcapacity. Such a targeted approach offers a clearer path to resolving the compute mismatch, while advancing Hong Kong’s ambitions as a world data hub and regional AI innovation center.

Contact the writer at grace@chinadailyhk.com