Executive summary

Two prizes from AI in Thai data centres
Thailand · THB bn a year, 2030 · bar = midpoint of range · rent = AI share of colocation revenue
AI tenants' rent, 2030THB 17-33 bnAI inside operations, 2030THB 1.8-6.7 bnof which cooling and power buyingTHB 0.6-2.6 bnBAR = MIDPOINT OF RANGE, THB BN A YEAR
Source: [Kira estimates] from [DC Byte 2025]; [BMI Jun 2026]; [LH Bank 2026]; [Bloomberg Jul 2026]

The answer in three lines

AI is the tenant now. AI took 28% of Bangkok capacity in early 2025, up from about 20% in 2024 [DC Byte 2025]; a Japanese firm is installing 4,696 B200 GPUs near Bangkok [Datasection Jun 2026].

AI as a tool is still a promise. AI cooling control cut cooling energy by up to 40% at Google [Google 2016]; we found no published result from a Thai hall.

The gain is energy, not headcount. In a 20 MW hall the cooling saving is THB 20-71 mn a year, against a payroll near THB 23 mn [Kira estimates].

Sell into the AI build now, and prove the in-hall gain in kilowatt-hours, which Thailand's new tariff and tax rules already pay for.
Bloomberg Jul 2026 · BMI Jun 2026 · Datasection Jun 2026 · DC Byte 2025 · Google 2016 · LH Bank 2026 · Kira estimates = KIRA analyst triangulation · full citations in Methodology and sources

Cooling, power buying and maintenance sit where data already exists and the money is largest; onboarding and rostering come last

Ten activities of a Thai data centre, scored on how much usable data already exists and how much value AI could move.

Data centre activities by data readiness and value at stake
Thailand · scores 1-5 · outside-in estimates · Oct 2026
Value, data to buildStart hereLeave for laterData rich, value smallHow much usable data already exists (1-5)Value AI could move (1-5)Cooling controlPower buying and load forecastingPredictive maintenanceCapacity planningCompliance reporting (PUE, water)Security monitoringDesign and digital twinConstruction managementCustomer onboarding and ticketsShift rostering
Source: [Kira estimates] from the evidence in Chapters 1 and 2

Why cooling leads

Energy is the cost. Power is about 70% of a Thai data centre's operating budget [Nation Aug 2026]; grid power cost THB 3.95 per kWh in May-Aug 2026 [LH Bank 2026] and a data-centre tariff of THB 5-6 is expected [Bloomberg Jul 2026].

The data is already logged. The first AI cooling models were trained on readings from thousands of existing sensors [Google 2016].

CaveatScores are our outside-in reading of public signals, not survey results [Kira estimates].
Start where the meters already are, in cooling and power, then maintenance; leave sales and rostering tools for later.
Bloomberg Jul 2026 = Bloomberg via Eco-Business, DC tariff · Google 2016 = Google/DeepMind, AI cooling · LH Bank 2026 = LH Bank Business Research · Nation Aug 2026 = The Nation, cost of the cloud · Kira estimates = KIRA analyst triangulation