Thailand's Energy Regulatory Commission is preparing a screening framework for data centres that is expected to take effect in the final quarter of this year, setting conditions on an industry the country has spent two years courting. Under the plans, data centres would pay THB5 to 6 per kilowatt hour against roughly THB3 for households, put up grid-access deposits proposed at THB4.5 million per megawatt, submit water management plans, install local energy storage, and face penalties if sudden demand increases force emergency generation.
Projects would also be screened on their electricity consumption, effect on grid stability, water use and economic contribution, and would have five to seven years to begin commercial operations or risk losing their allocated capacity.
The scale of what Thailand is trying to manage explains the caution. Data centres and AI could eventually require up to 30,000 megawatts in the country, against a Power Development Plan targeting 51 gigawatts of renewable capacity, while foreign investment applications reached THB1.01 trillion in the first quarter of 2026, with THB873.7 billion of that in digital projects.
AI Is Already Running Parts Of The Grid
The same technology creating the demand is being used to manage it. At a sustainability finance event in Bangkok in September, energy executives described using AI to design renewable projects before committing capital, forecast how much power variable sources will deliver, and balance supply across the grid.
Ratchaneewan Pulnil, country manager for Thailand at Gurīn Energy, said technical analysis that once took weeks now takes hours or days, and that the investment benefit is tangible. EGAT, the state generating authority, is deploying AI forecasting models to predict renewable output and handle the volatility that comes with it.
Those tools matter more as the share of solar and wind rises, because a grid with large amounts of variable generation has to be dispatched far more actively than one running on gas and coal.
The Measurement Problem
The limitation is what the grid can see. Fewer than 300,000 of Thailand's roughly 30 million electricity meters are smart meters, under 1% of the total, according to Pattanan Amatanon of Forth Corporation, speaking at a forum in August.
Forecasting models are only as good as the data feeding them, and a distribution network with almost no real-time visibility at the edge cannot be optimised in the way operators in Europe or Australia now take for granted. Industry figures at that forum put the value of smart grid development at over THB30 billion a year, including THB28 billion in line losses and maintenance that better monitoring could avoid.
Permits Move Slower Than Factories
The bottleneck most often cited by executives is administrative rather than technical. Nopadej Karnasuta of B.Grimm Power described a timeline mismatch in which permits take six years while factories take three to six months to build.
Akarin Prathuangsit of WHA Utilities and Power argued that investment decisions now turn on access to clean energy rather than labour costs, and warned that delays risk pushing investors to other countries for good. That is a direct concern in a region where Malaysia, Singapore and Indonesia are competing for the same AI infrastructure.
Thailand's answer is to open up power purchasing. The government has expanded direct power purchase agreements beyond a 2,000 megawatt pilot that was limited to data centres, so industrial buyers can buy renewable power directly from generators and move it across state networks for a regulated fee. Wheeling charge regulations are due to be finalised this year, with the first deals expected in 2027 and a smart control centre planned for 2028.
What This Means For Operators Choosing Sites
For companies deciding where to put AI capacity in Southeast Asia, Thailand now presents a clearer but more demanding proposition. Power is available and the renewable pipeline is substantial, but the terms are becoming explicit: a higher tariff, a capital deposit against grid access, commitments on water and storage, and a deadline to actually build.
Operators should price those conditions into site comparisons rather than treating announced capacity as available capacity. They should also check where direct power purchase agreements stand, since the ability to contract renewable supply directly affects both cost and the clean energy claims a customer can make.
The regional picture matters too. Oracle's reported lease of compute capacity to Tencent across Southeast Asian data centres is one sign of how much demand is heading to this region, which tightens competition for power, land and grid connections everywhere in it.
Expect Conditions, Not Open Doors
Thailand's approach reflects a wider shift. Governments that spent recent years attracting data centres with incentives are now attaching requirements, because the power these facilities consume competes with households and existing industry.
Operators planning deployments through 2027 and beyond should expect screening, tariffs that reflect the true cost of supply, and obligations on water and storage to become standard across the region rather than a Thai peculiarity.
The Grid Decides Who Hosts The AI Boom
Thailand has the renewable ambition, the investment interest and increasingly the AI tools to run a more complex grid. What it does not yet have is the metering, the permitting speed or the transmission capacity to match, and those are slower to fix than announcing a target.
The next 18 months will show whether the country can close that gap. If the wheeling rules land on time, direct power agreements start in 2027 and smart metering finally scales, Thailand becomes a serious option for AI infrastructure. If permits still take six years, the capacity will go to whichever neighbour moves faster, and the same pressure will keep moving down the stack, as it has in the fibre networks carrying AI's uplink surge behind 5G towers.
