Malaysia is not too hot for dry cooling. Across 11 data-centre districts, outside air passed 32°C for 13 to 142 hours a year from 2016 to 2025, under 2% of the year, and the hottest hour in ten years was 33.8 to 36.1°C (ERA5 via Open-Meteo).
It is humid rather than extreme: the 99th-percentile wet-bulb temperature was 26.3 to 26.9°C in every district. Rain is frequent, with the longest dry spell 7 to 17 days and the driest 60 days still bringing 63 to 165 mm.
- 13–142 h/yr
- Hours above 32°C, by district, 2016–2025 (ERA5)
- 33.8–36.1 °C
- Hottest hour in ten years, by district
- 26.3–26.9 °C
- 99th-percentile wet-bulb, by district
- 11
- Districts in the dataset, free to download
Is Malaysia too hot for dry cooling?
No. A dry cooler rejects heat to outside air, so what matters is how often the air is hotter than the cooling water. From 2016 to 2025 the 11 districts passed 32°C for 13 to 142 hours a year, and 34°C for 0 to 15 hours. Even the hottest district, Nilai, was above 32°C for 142 hours a year, 1.6% of the 8,760 hours.
Is liquid cooling needed for AI data centres in Malaysia?
No national rule requires liquid cooling. Johor, since 10 March 2026, considers only Tier 3 and 4 air-cooled designs and refuses high-water-demand designs (Malay Mail). On the climate, warm-water liquid cooling suits Malaysia: with cooling water at 40°C or more and dry coolers running about 6°C above outside air, heat can go to the air whenever it is below about 34°C. Every district’s hottest hour in ten years was 33.8 to 36.1°C, so this works almost all year. That is our engineering reading of the data, not a design standard.
How many hours a year is each district above 30°C or 32°C?
| District | State | Above 26°C | Above 30°C | Above 32°C | Above 34°C | Median | 99th pct | Hottest hour | Wet-bulb 99th pct / max |
|---|---|---|---|---|---|---|---|---|---|
| Kulai / Sedenak | Johor | 3,738 | 443 | 33 | 2 | 25.6°C | 31.4°C | 35.5°C | 26.9 / 28.7°C |
| Iskandar Puteri / Nusajaya | Johor | 5,106 | 516 | 48 | 2 | 26.5°C | 31.6°C | 35.8°C | 26.9 / 28.5°C |
| Pasir Gudang | Johor | 5,017 | 633 | 63 | 3 | 26.4°C | 31.8°C | 36.0°C | 26.9 / 28.1°C |
| Pengerang | Johor | 5,479 | 246 | 13 | 0 | 26.6°C | 30.9°C | 34.2°C | 26.3 / 27.8°C |
| Cyberjaya | Selangor | 4,469 | 834 | 112 | 9 | 26.1°C | 32.2°C | 35.7°C | 26.8 / 28.3°C |
| Banting | Selangor | 5,056 | 677 | 34 | 0 | 26.5°C | 31.5°C | 33.8°C | 26.8 / 28.2°C |
| Nilai | Negeri Sembilan | 4,289 | 910 | 142 | 15 | 26.0°C | 32.5°C | 36.0°C | 26.7 / 28.1°C |
| Port Dickson | Negeri Sembilan | 5,329 | 618 | 47 | 1 | 26.7°C | 31.6°C | 35.0°C | 26.9 / 28.4°C |
| Kulim | Kedah | 3,921 | 872 | 138 | 11 | 25.6°C | 32.5°C | 36.1°C | 26.8 / 28.3°C |
| Bayan Lepas | Penang | 5,762 | 312 | 18 | 0 | 26.8°C | 31.0°C | 34.5°C | 26.8 / 29.0°C |
| Kota Samarahan | Sarawak | 3,638 | 730 | 109 | 4 | 25.5°C | 32.2°C | 35.2°C | 26.9 / 28.2°C |
ERA5 reanalysis (Copernicus) via Open-Meteo, hourly, 2016–2025. Analysis: Croutons Research.
The coast is milder at the peaks: Pengerang (13 hours above 32°C), Bayan Lepas (18) and Banting (34). The inland belt runs hottest: Nilai (142), Kulim (138) and Cyberjaya (112).
Download all 11 districts as CSV, free to reuse with credit (CC BY 4.0). Credit: ERA5 / Copernicus via Open-Meteo.
Can rainwater cover data-centre cooling water?
For peak-hour evaporative top-up, on our estimate, yes. Rain by district:
| District | Rain, average | Rain, driest year | Longest dry spell | Driest 60 days |
|---|---|---|---|---|
| Kulai / Sedenak | 3,008 mm | 2,249 mm | 7 days | 150 mm |
| Iskandar Puteri / Nusajaya | 2,873 mm | 2,354 mm | 9 days | 135 mm |
| Pasir Gudang | 3,274 mm | 2,578 mm | 9 days | 136 mm |
| Pengerang | 2,600 mm | 1,962 mm | 9 days | 63 mm |
| Cyberjaya | 3,308 mm | 2,439 mm | 14 days | 124 mm |
| Banting | 3,592 mm | 2,621 mm | 10 days | 117 mm |
| Nilai | 3,110 mm | 2,390 mm | 11 days | 123 mm |
| Port Dickson | 3,013 mm | 2,161 mm | 11 days | 119 mm |
| Kulim | 3,822 mm | 2,127 mm | 11 days | 70 mm |
| Bayan Lepas | 3,533 mm | 2,207 mm | 17 days | 115 mm |
| Kota Samarahan | 4,043 mm | 2,823 mm | 12 days | 165 mm |
Kulai / Sedenak
Iskandar Puteri / Nusajaya
Pasir Gudang
Pengerang
Cyberjaya
Banting
Nilai
Port Dickson
Kulim
Bayan Lepas
Kota Samarahan
ERA5 reanalysis (Copernicus) via Open-Meteo, hourly, 2016–2025. Analysis: Croutons Research.
The calculation (an estimate, with assumptions)
- Assume a 20 MW hall rejects all its heat by evaporation only in peak hours, about 150 hours a year (close to the highest count above 32°C, Nilai’s 142).
- Heat: 20 MW × 150 h = 3,000 MWh.
- Evaporating one litre of water absorbs about 2.45 MJ, or 0.68 kWh.
- Water: 3,000,000 kWh ÷ 0.68 kWh/L ≈ 4,400 m³ a year, before blowdown.
- Rain on a 2-hectare roof (20,000 m²) in Kulai’s driest year (2,249 mm): 20,000 × 2.249 ≈ 45,000 m³.
So a roof that size collects about ten times the peak-hour top-up even in a dry year, if the water can be stored between storms. Storage is the constraint: Pengerang’s driest 60 days brought only 63 mm. These are our estimates, not measurements; size real systems from MetMalaysia station data.
How are data centres cooled without water?
With dry coolers or air-cooled chillers, which reject heat to outside air through fins and fans instead of by evaporation. They use more fan and compressor energy on hot hours, which shows up in PUE. Liquid cooling at a high water temperature widens the hours they can run without compressors. Water per AI answer depends on this choice: see energy, water and carbon per answer.
Method. ERA5 reanalysis via Open-Meteo, hourly, 2016 to 2025, on a grid of about 9 to 25 km. ERA5 smooths peaks, so treat hour counts as lower bounds, especially in urban Cyberjaya. Use MetMalaysia station data or ASHRAE design conditions for engineering. ERA5 rainfall can read high against rain gauges.
What is not public
We could not confirm these from a published source as of 10 November 2026. Do not assume them:
- Measured PUE or WUE for named Malaysian sites
- MetMalaysia station records as open data at hourly resolution
- Design conditions used by individual operators
Croutons Research publishes open data and methods on AI data centres in Malaysia. See the open method for measuring AI work per megawatt.
Sources
- Open-Meteo Historical Weather API (ERA5 reanalysis, Copernicus)
- Malay Mail, Johor refuses high-water designs (10 Mar 2026)
Last checked 10 November 2026