Hong Kong’s expanding AI capacity brings higher rack densities, liquid-cooling requirements and stricter energy disclosure into a single investment case centred on resilience, operating cost and capital discipline.
Hong Kong’s AI data-centre build-out enters a higher-density phase as CETA System Co., Limited examines the operating economics behind the capacity expansion. HKSTP and SenseTime have agreed to develop the city’s largest domestically produced AI data centre, targeting more than 40,000 petaflops by the end of the decade. The first phase is planned before year-end, using domestically produced GPU clusters for model training, inference and large-scale application deployment. The Science Park already hosts more than 500 AI companies employing over 5,000 AI specialists.
Private-sector capacity also expands through Equinix’s $136.1 million commitment to HK6, its sixth International Business Exchange in Hong Kong. It represents the company’s largest single investment in the city over the preceding decade. The first phase is designed to provide 1,000 cabinets, rising to 3,550 at full build-out, while the facility is configured for liquid cooling and high-density AI workloads. Hong Kong Internet Exchange traffic increased by nearly 50% over the preceding year, adding commercial context to the investment.
GPU heat loads define the engineering constraint as chipsets exceed 1,000 watts and approach 2,000 watts or more in AI workloads. These loads push rack densities towards 100 kW per cabinet, while conventional air cooling reaches a practical ceiling at about 70 kW per rack. Global Switch has deployed 30 MW of liquid-cooled capacity in Tseung Kwan O, while direct-to-chip cooling and rear-door heat exchangers are in active use. Deployment data indicate that liquid cooling can cut energy consumption by up to 30% against comparable air-cooled configurations.
Lee Tsz-Hin, Chief Executive Officer of CETA System Co., Limited, describes this as “a capacity-matching problem with direct consequences for uptime, energy use and capital efficiency”. At high density, cooling flow must be aligned with the heat output of each rack. Localised hotspots and sustained peak loads can otherwise push conventional thermal systems beyond stable operating limits.
Downtime makes the financial downside quantifiable, with industry survey data estimating costs at $10,973.4 per minute per rack. Thermal incidents carry detection and recovery costs above $536,600.1 per event. Over the periods covered by the surveys, thermal issues account for 15% of recorded root causes across critical infrastructure failures, while about 80% of critical-infrastructure organisations report operational technology downtime losses between $132,000 and $6.6 million per incident. These exposures strengthen the case for condition-based monitoring and predictive maintenance alongside cooling investment.
For CETA System, the central operating question is whether operators can obtain that visibility without replacing assets across multi-generation estates. Building-management and data-centre infrastructure management platforms can be linked across BACnet, Modbus, SNMP and REST APIs, allowing analytical models to compare equipment behaviour, consumption history, cooling setpoints and chiller performance. Lee calls the advisory-first model “the practical route to operator trust, because recommendations remain subject to human review, hard constraints and failsafe modes”. Autonomous control is not the objective; operator oversight remains the governing principle.
Regulatory obligations add a second operating constraint as Hong Kong’s amended Buildings Energy Efficiency Ordinance extends coverage to stand-alone data centres and converted industrial space. The expanded coverage captures more than 75% of local data-centre gross floor area. Energy-audit intervals fall from 10 years to five, alongside mandatory disclosure of efficiency coefficients, control systems and equipment service life. The framework is estimated to deliver an additional 500 million kWh in annual savings by the middle of the next decade.
For operators with cross-border exposure, the EU Energy Efficiency Directive establishes public reporting through a European database for qualifying data centres. The threshold is installed IT power demand of at least 500 kW, and the required disclosures cover energy-performance indicators. Scope 1, 2 and 3 emissions disclosure further makes asset-level data collection a governance requirement, not merely an operational preference.
That combination changes the investment lens by tying recurring operational-intelligence revenue to real-time monitoring and data extraction from legacy equipment. Predictive analytics for UPS systems, generators and chillers forms part of the same revenue model across distributed facilities. “Recurring value comes from measurable efficiency and reliability gains rather than speculative capacity alone,” Lee observes, linking the commercial case to lower operating cost, improved resilience and reduced downtime risk.
Hong Kong’s AI capacity push therefore turns on the interaction between headline compute, thermal resilience, integration and energy transparency. Facilities approaching 100 kW per cabinet must align liquid cooling, vendor-agnostic integration, predictive maintenance and auditable energy performance as a single capital-allocation discipline. CETA System views that convergence as the basis for infrastructure that can support AI growth without separating technical resilience from compliance exposure and long-term operating economics for operators and institutional investors alike.
About CETA System
CETA System Co., Limited is a Hong Kong-incorporated technology company founded in 2017 to serve data-centre infrastructure. Its vendor-agnostic artificial-intelligence platform combines HVAC and chiller-plant energy optimisation with predictive maintenance for UPS systems, generators and chillers. The advisory-first platform integrates with existing building-management and DCIM environments for colocation, enterprise and hyperscale operators across Asia-Pacific and other markets.
- Website: https://cetasystem.com
- Registered business: CETA System Co., Limited (Hong Kong BRN 67731517; CRN 2533166)
