Where to build: a better data-center site than Chiang Mai

Research note, October 27, 2026, Alexander Grabowski (B4L). Sources retrieved October 27, 2026; figures are cited and our estimates marked.

This note looks for a data-center site with room for solar and possibly wind. It compares ten candidates with Chiang Mai and with the Eastern Economic Corridor (EEC), using our screening model, model/site_energy.py in the repository. The model matches hourly solar and wind output for 2019 to 2023 against 1 MW of IT at a PUE of 1.2, where PUE is total facility power divided by IT power. The building itself is described in option B. Money is in THB, with USD at 33 THB per USD, the B4L planning rate.

In short

• Thailand's inland wind is in Nakhon Ratchasima (Korat) and Chaiyaphum. A JGSEE/KMUTT study for NSTDA and EGAT ranks Huai Bong (Dan Khun Thot, Korat) second of all sub-districts for wind at a capacity factor of 20% or more, while the whole Upper North, Chiang Mai included, has 0.02 GW of such potential.[307] Farms in the belt report or expect about 19 to 21%.[308][309]
• Solar barely separates the sites. Our model gives 1,413 to 1,499 kWh per kWp a year across the candidates, against 1,418 at Mae Hia in Chiang Mai, so land, wind, grid, hazards, and water decide.
• Per MW of IT, 5 MWp of solar with a 10 MWh battery covers about 62 to 65% of the load hour by hour at 3,100 to 3,400 THB per MWh (USD 94 to 103), below the all-grid price of 3,810 THB (USD 115). With 10 MWp and 20 MWh, coverage rises to about 95% at roughly the grid cost, on about 100 rai of land (our estimates).
• Surplus power cannot be sold, because net billing stops at 5 kW per meter[310] and the Direct PPA pilot needs 50 MW of IT per building.[311] The rest of the power is best bought as UGT2 green power at 4.04 to 4.40 THB/kWh before VAT on 10-year contracts.[312][313]
• The candidates in the Northeast and the central region lie in no seismic zone, while Chiang Mai is in Zone 3.[314] Outside the EEC a high-efficiency data center falls in BOI group A2 with 8 years of tax exemption, against 5 years (group A3) inside it.[135]
• Our ranked shortlist, as a judgment, is (1) Sikhio and the Lam Takhong corridor (Korat), (2) Thep Sathit and Sap Yai (Chaiyaphum), (3) Muak Lek (Saraburi), (4) Khon Kaen, and (5) the 304 Industrial Park (Prachinburi).
• Two risks remain. On September 4, 2026, the data-center policy board paused 166 projects while it writes new rules,[315] and Korat's main reservoir held 16% in May 2025.[316]

What the site has to offer

On BOI terms, a high-efficiency data center (PUE 1.3 or lower, at least 2 MW of IT) is group A3 inside the EEC and A2 elsewhere.[319] Chaiyaphum, Ubon Ratchathani, Sa Kaeo, and Mukdahan are among the 20 lowest-income provinces; Nakhon Ratchasima, Khon Kaen, Saraburi, and Prachinburi are not.[319] In those provinces, A1 and A2 projects get a 50% cut in corporate income tax for five years after the exemption instead of three extra years, and the guide's table adds a double deduction of transport, electricity, and water costs for 10 years.[319] For a business whose biggest running cost is electricity, that deduction may be worth more than the tax holiday (our reading, to confirm with BOI). BOI is also mapping areas outside the EEC that have enough transmission capacity and are safe from floods and earthquakes.[322]

Chiang Mai falls short on most of these criteria. It is in seismic Zone 3, the highest band.[314] In October 2024 the Ping River rose about 1.08 m above the city's flood defenses.[29] The North had 129 days above the PM2.5 standard in 2024 and a 24-hour peak of 218 µg/m³, against 113 days and 129 µg/m³ in the Northeast.[321] Its wind is negligible: 0.02 GW for the Upper North,[307] and about 1% capacity factor in our model. Its solar is within about 5% of the best candidate, and it is cooler, with about 1,870 hours a year below 20 °C against about 670 at Sikhio (our model).

Wind: where Thailand has it

The older DEDE map put utility-grade wind only on the southern Gulf coast and at Doi Inthanon.[323] The JGSEE/KMUTT study compared four maps at 100 m and finds 17 GW of potential at a capacity factor of 20% or more with low-wind turbines, and 5 GW with conventional ones.[307] By region, the southern Gulf coast leads with 7.64 GW, the Lower Northeast (including Korat) has 1.66 GW, the Upper Northeast (including Chaiyaphum) 0.73 GW, and the Upper North 0.02 GW.[307]

The ranking comes from JGSEE/KMUTT, Table 8.[307]

Theppana recorded plant factors of 18.8 to 21.1% from 2014 to 2017,[308] and RATCH expected each 103.5 MW West Huay Bong farm to deliver 175 million kWh a year,[309] about 19% by our arithmetic. The ERC assumes 24% for wind in the UGT2 portfolios[169] and BNEF 20% for 2024.[337] Our model, with a generic 4.2 MW low-wind turbine scaled to that range, gives about 21% for the Korat and Chaiyaphum cells, 14% for Muak Lek, and 1% for Chiang Mai (our estimate). Its cells are about 50 by 60 km, so the model ranks regions; only a measurement mast can size a turbine.

Korat and Chaiyaphum are the wind belt. The operating farms listed above in these two provinces add up to about 1.2 GW (our sum), on a strip about 90 km long from Lam Takhong north to Sap Yai (our measurement from published coordinates).[338][339]

The candidates compared

Seismic zones follow the province lists of the 2021 regulation;[314] provinces it does not list are marked none. C1 and C2 share one wind cell. Treasury appraisals span everything from farmland to town centers (100 to 130,000 THB per square wah in Nakhon Ratchasima), so the table shows only listings.[361]

C1 Sikhio has farmland in a green (rural and agricultural) zone listed at about 300,000 THB per rai.[340] Korat's chamber of commerce pitches Sikhio's pumped storage, wind, and solar to data-center investors.[362] Suranaree University of Technology (about 17,000 students in 2020) is in Korat,[363] and Gorilla Technology says it has secured about 100 rai in Korat for a 200 MW AI campus (company claim).[364]

C4 Thep Sathit and Sap Yai has the highest solar yield, the densest wind farms, the cheapest land found, and the BOI terms of a lowest-income province. It is further from Bangkok and from a university, though, and no fiber route was found.

C5 Muak Lek is the closest wind area to Bangkok. INET's largest data center is in the province,[348] and WHA is adding almost 3,000 rai of estate land for future data centers.[349]

C7 Khon Kaen brings talent: KKU runs an H100 AI cluster and a BSc in AI,[352] and INET is building a 2,500 million THB data center there for the second quarter of 2027.[353]

C9, the 304 Industrial Park, offers estate power and water. In the October 2013 floods its operator expected the park to stay dry at 20 m above sea level (lowest point 13 m) behind a raised barrier.[357] It has no wind.

Five candidates did not make the shortlist. C3 Huai Bong is remote, and C4 offers the same wind belt with better BOI terms. C2 Suranaree has estate prices, C6 Lopburi is hot and lies on low plains, C8 Sirindhorn is far away and prone to floods, and C10 Khao Kho is forest land with a single 115 kV line.

How much of the power on-site solar, batteries, and wind can cover

Our model runs a flat 1.2 MW load (10,512 MWh a year) against hourly PVGIS solar and NASA POWER wind data, with a 4-hour battery at 90% round-trip efficiency. "Hourly" coverage is the share of the load met in the same hour by on-site plant, directly or through the battery; "annual" coverage is generation divided by use.

The figures are our estimates from the model. Land is taken at 10.1 rai per MWp from the ERC filing[317] and at 1.5 rai per MW of wind from the operating footprint of Hanuman 9.[365]

Annual matching flatters solar. At Sikhio, 10 MWp makes 138% of the year's use yet covers only 42% of the hours. Google measures each data center's hourly match with carbon-free supply,[293] and Microsoft has committed to matching 100% of its use, 100% of the time, with zero-carbon purchases by 2030.[366] Ember, using hourly profiles, estimates that about a third of ASEAN data-center demand in 2030 could be met by solar and wind without batteries.[367][368] The battery does the work here: 20 MWh per MW of IT lifts 10 MWp from 42% to about 95%. Wind fills the nights. One turbine alone covers about half the hours in the belt, and almost none at Chiang Mai.

The catch is the surplus. The net-billing scheme approved on April 29, 2026 pays 2.20 THB/kWh for rooftop surplus, but only up to 5 kW per meter,[310] and commercial and industrial systems are normally built for zero export (installer guidance).[369] Our model therefore counts surplus as lost. At 10 MWp with 10 MWh, 44% of Sikhio's solar output is thrown away (our estimate).

What it costs

The ranges that follow run from the low to the high solar cost. The cheapest mix at every candidate is 5 MWp with 10 MWh per MW of IT. It covers 62 to 65% of the load hour by hour at 3,083 to 3,424 THB per MWh of load, against 3,810 THB from the grid alone. At Sikhio it costs 3,240 to 3,356 THB (USD 98 to 102), which saves about 4.8 to 6.0 million THB (USD 145,000 to 182,000) a year per MW of IT after capital charges, for 171 to 184 million THB (USD 5.2 to 5.6 million) of plant and land. More coverage costs little more. With 7.5 MWp and 20 MWh, coverage reaches 88 to 92% at 3,189 to 3,664 THB, and with 10 MWp and 20 MWh it reaches 95 to 96% at 3,591 to 4,127 THB, about the grid cost. At Sikhio that larger plant and its land cost 342 to 368 million THB (USD 10.4 to 11.2 million), roughly as much as the building itself (all our estimates).

The sites differ little. Land is charged only where a price was found, and with one land price everywhere the cheapest mix spans only about 3,070 to 3,190 THB per MWh (at the low end of the solar cost) across all twelve sites (our extra run).

Wind does not pay on its own at this scale. One 4.2 MW turbine costs about 225 million THB (USD 6.8 million) at the BNEF farm price and is three and a half times the size of the load, so 28% of its output at Sikhio is lost. No mix with a turbine at any candidate comes below about 4,050 THB per MWh. At 2 MW of IT sharing one turbine, an extra run gives 86% coverage with 5 MWp and 10 MWh per MW of IT at 3,583 to 3,699 THB. That is slightly worse than a bigger battery without a turbine, since 7.5 MWp with 20 MWh gives 90% at 3,409 to 3,583 THB (our estimates). At this scale, then, wind is a hedge and does not lower the cost.

The cheapest mix beats the grid in every case except on EEC land, where the model's best choice is no on-site plant at all. The battery price decides whether near-full coverage beats the grid too. Buying the uncovered 37% as UGT2 instead of grid power adds about 190 to 330 THB per MWh, so 5 MWp with 10 MWh plus UGT2 is green on an annual basis at about 3,430 to 3,690 THB per MWh at Sikhio, still below the all-grid price (our estimate, assuming UGT2 can cover part of a load). If the separate data-center tariff of 5 to 6 THB/kWh now under consideration is adopted, every on-site mix gains.[134]

Rules for generating power on site and buying green power

Direct PPA is moving but does not fit this project. The National Energy Policy Council extended it to industrial users on July 15, 2026, with the final wheeling charges still outstanding in August,[165] and the minister said on August 28 that the 2,000 MW cap had been lifted.[382] Even so, the draft medium-voltage adders total 2.55 THB/kWh, which with solar at about 2.16 THB/kWh (the minister's figure)[382] puts delivered power near 4.7 THB/kWh, above UGT2 (our arithmetic). UGT2 has also fallen from 4.5475 THB/kWh in its 2024 design.[169] A 1.2 MW load is a PEA large general service customer (1,000 kW and above),[25] which is the class of large user that UGT2 is meant for.[313]

Three policy moves matter. The ERC consulted from September 19 to October 18, 2026 on a Data Center Electricity Distribution License, whose applicants would file a supply plan by source, self-generation included, and an RE100 target year.[383] On September 4, 2026 the policy board paused 166 projects, 49 under construction and 117 awaiting permits, and will reconsider regulating data centers above 2 MW of demand as factories.[315] And data-center requests to PEA and MEA total about 35,000 MW against about 8,800 MW in the draft PDP2026, while EGAT studies backup charges for self-generators.[384]

Grid, water, and hazards

Grid

EGAT's June 2026 tender covers a new 500/230 kV Nakhon Ratchasima 4 substation and expansions of the 500 kV Chaiyaphum 2 and 230 kV Nakhon Ratchasima 3 substations,[342] and in May 2025 EGAT awarded a 1.46 billion THB voltage-support (STATCOM) contract at Chaiyaphum 2.[385] The ERC approved a 500 kV route from Korat 4 to Wang Noi through Saraburi to carry renewable power from the Northeast to Bangkok.[347] EGAT's Lam Takhong pumped-storage plant in Sikhio has 1,000 MW.[341] Substation sites were not published, and outside estates a 115 kV customer co-invests in the line.[290]

Water

Lam Takhong held only 16% on May 3, 2025,[316] and Korat's chamber of commerce plans talks with the provincial irrigation office about supplying data centers.[362] The 304 estate offers 344,000 m³ a day,[356] and WHA has already contracted 30 million m³ a year to data-center customers.[386] A closed-loop design with dry coolers, as in option B, keeps the need small (our judgment).

Floods

Floods in October 2010 hit 21 provinces, including Nakhon Ratchasima, Chaiyaphum, Saraburi, Prachinburi, and Khon Kaen.[344] In September 2019, 994 villages in 13 districts of Ubon Ratchathani flooded,[355] and in 2011 seven industrial estates stood under up to 3 m of water.[387] The Sikhio plot lies at about 260 m.[340] We did not retrieve flood maps for individual parcels.

Earthquakes

The 2021 regulation[320] puts Chiang Mai and other northern and western provinces in Zone 3, and Bangkok and the lower central provinces in Zone 2. Nakhon Ratchasima, Chaiyaphum, Saraburi, Khon Kaen, and Prachinburi are not listed.[314]

Haze and heat

In 2024 the North had 129 days above the PM2.5 standard, the Northeast 113, and the Central region 101.[321] On February 23, 2026 the Northeast had the day's highest reading, 84.9 µg/m³, with Khon Kaen, Saraburi, and Prachin Buri among the provinces to monitor.[388] Korat is warmer than Chiang Mai (a mean of 26.3 against 23.3 °C in our model), which matters little with warm-water liquid cooling.

Ranked shortlist

We checked the ranking against the model. The model keeps the top two on top for wind (21% each) and puts Thep Sathit slightly ahead on cost (3,118 to 3,234 against 3,240 to 3,356 THB per MWh for the cheapest mix), thanks to more sun and cheaper land. We keep Sikhio first for grid, distance, and talent, which the model does not weigh. The model would swap places 3 and 4: Muak Lek has the weakest wind cell on the list (14%) and the highest cost of any candidate (3,308 to 3,424 THB), while Khon Kaen's cell shows 18% at 3,084 to 3,200 THB. We keep Muak Lek third for its short distance to Bangkok and its data-center neighbors, and because Khon Kaen's 18% comes from a coarse cell with no ranked sub-district or farm behind it. At one land price the shortlist sits within about 4% on energy cost, so wind, grid, water, and people set the order.

What this means for the investor

What follows is our judgment.

The search should focus on the Korat and Chaiyaphum belt. A parcel of 100 to 150 rai per MW of IT carries 5 to 10 MWp of solar and a 10 to 20 MWh battery, which covers roughly two thirds to nearly all of the power hour by hour at or below the grid price, and UGT2 covers the rest. Chiang Mai's sun is nearly as good, but its earthquake zone, floods, haze, and lack of wind count against it.

Solar should be sized to the battery rather than to the annual total. Because surplus cannot be exported, panels beyond about 5 MWp per MW of IT pay only with more storage, and the battery price decides whether 95% coverage beats the grid.

Wind on site makes sense only with a partner. One turbine does not pay at 1 MW of IT. At 2 MW or more, it is worth testing if a developer in the belt (WEH, EGCO, Energy Absolute, or Gulf) will build or share a turbine, a mast confirms the wind, and the land title allows it.

CMU ITSC remains the pilot. Option A stays the first step, and the choice of site matters once anchor load nears the 2 MW BOI threshold.

The policy pause and water need watching. The September 2026 pause and the proposed license could change where and how data centers connect, and water is the open point in Korat.

Open questions