HK triples data pipeline as AI sparks $3t supercycle

AI inference requirements are forecast to surpass training as the dominant infrastructure demand by 2027.

Hong Kong's global data centre sector is entering a high-friction "infrastructure investment supercycle," with total capacity forecast to nearly double to 200 GW by 2030.

This expansion is expected to require $3t in total investment over the next five years, including $1.2t in real estate asset value and $870 billion in new debt financing, according to JLL.

 While the city completed 1.8m sq ft of space (150 MW load) between 2022 and 2025, JLL reports a massive 4.8m sq ft pipeline scheduled for delivery between 2026 and 2029. This represents a nearly 3x surge in supply, designed to capture a share of the $1t hyperscalers are committing to infrastructure through 2026.

The surge comes amidst a "perfect storm" of global supply constraints, with grid connection delays now reaching four years in some markets. To bypass these bottlenecks, Hong Kong is accelerating land supply via the Sandy Ridge Data Facility Cluster and the San Tin Technopole. These government-backed clusters aim to support AI training facilities, which demand 10x the power density of traditional centers and command lease premiums of up to 60%.

JLL predicts a critical inflection point in 2027, as AI inference workloads surpass training as the dominant market requirement. With extensive submarine cable connectivity and low exposure to natural hazards, Hong Kong is leveraging supportive policy to absorb the next wave of sovereign and enterprise digital transformation.