Constraint-First Siting

Southeast Asia's AI buildout is being sited by its substrate, not its chequebook. Call it constraint-first siting: power, water, tropical cooling loads, and political permission decide where the work lands; capital routes itself to whatever the substrate will bear.

Turner & Townsend's 2026 projection of $30 billion in regional data-centre investment by 2030, paired with the U.S.-ASEAN Business Council's 20%-annual demand forecast through 2028, reads like a capital-led boom. It is not. It is a backlog of capacity chasing a finite physical and regulatory floor.

Singapore is the cleanest case. ARC Group's 1.4% vacancy rate and ~1GW of operational capacity show a market full on purpose — land scarcity, imported power, and green rules used as a filter rather than friction. Microsoft's $5.5 billion Singapore commitment, AWS's $9 billion expansion, and Google's regional footprint all land in the same small, expensive, strategic node. The substrate admits one tier of tenant and excludes the rest.

Indonesia, Malaysia, Thailand, Vietnam, and the Philippines hold the volume — over 2,000 operational sites and 1,000+ in planning (Ember, 2026) — but tropical year-round heat, grid lags, and shifting data-localisation rules do the actual siting. The mechanism is portable: the binding substrate admits the workload or it doesn't, and the chequebook arrives afterwards.

The map of where AI infrastructure will actually arrive is drawn in megawatts, cooling degrees, and ministerial sign-off — not in announced capex. The grid writes the cheque now.