Data Centre PUE in Malaysia: Targets, Measurement, Reality

Power Usage Effectiveness (PUE) is the total energy a data centre consumes divided by the energy its IT equipment consumes. A PUE of 1.5 means that for every kWh reaching a server, another 0.5 kWh goes to cooling, UPS losses and everything else. Since 30 December 2024, data centre PUE in Malaysia has stopped being a marketing number and become a compliance one: the MIDA/MITI Guideline for Sustainable Development of Data Centre ties DESAC tax incentives to design PUE targets of 1.4, 1.6 or 1.7 depending on facility category, declared per ISO/IEC 30134-2 as an annualized average. That last clause is the part most design brochures skip, and it is where this article spends most of its time.

What is PUE? The formula Malaysian regulators now use

PUE = EDC / EIT: total annual data centre energy (kWh) divided by annual IT equipment energy (kWh), as defined in ISO/IEC 30134-2, the standard the MIDA guideline explicitly references.

Worked example: a facility draws 26,280 MWh from TNB in a year. Its IT equipment (servers, storage, network) consumes 17,520 MWh of that. PUE = 26,280 / 17,520 = 1.5. The remaining 8,760 MWh went to chillers, CRAHs, pumps, UPS conversion losses, lighting and the office floor. A perfect (and physically unreachable) PUE is 1.0. Lower is better, and every 0.1 of PUE on a 20 MW IT load is roughly 17.5 GWh a year of overhead energy.

Two things regulators care about that vendors often blur:

- It is an annual ratio, not a snapshot. A chilled-water plant at 2 am in December looks nothing like the same plant at 3 pm during a February dry spell.
- The denominator is measured IT energy, not nameplate. You need a metered boundary between IT load and facility load, which we cover below.

Malaysia's official PUE targets: 1.4, 1.6 or 1.7 depending on your category

Appendix 1 of the MIDA guideline sets design PUE targets by facility class for DESAC tax-incentive applicants:

- Hyperscale data centres (high voltage, 132 kV and above; capacity above 21.25 MW): design PUE 1.4 and below.
- Purpose-built colocation, built after 2020, above 21.25 MW: design PUE 1.6 and below.
- Pre-2020 colocation, converted-building colocation, and enterprise/captive data centres (0.85 to below 21.25 MW at 33 kV or 11 kV): design PUE 1.7 and below.

Baselines "need to be established" for each facility, and targets are validated once the baseline is set, so you cannot claim compliance without measured data. DESAC applications received by MIDA until 31 December 2027 are subject to these PUE, WUE and CUE conditions.

Separately, MCMC's Technical Code for Green Data Centres reportedly sets 1.9 as a minimum PUE requirement and below 1.5 to qualify as green. We have only seen those figures in secondary legal summaries, so verify against the code itself before designing to them. The direction of travel, however, is unambiguous.

Is a data centre PUE of 1.4 realistic in a tropical climate?

Achievable, but only for well-run hyperscale facilities, and it puts Malaysia's target below what most of the world actually delivers. The Uptime Institute's 2025 global survey puts the weighted average annual PUE at 1.54, essentially flat for the sixth consecutive year. The industry fell fast from 2.5 in 2007 to 1.65 in 2014, then plateaued. The easy wins are gone.

Geography makes it harder here. Uptime's regional analysis found Europe averaging 1.46 and US/Canada about 1.51, while Asia-Pacific runs higher, around 1.68 by one industry estimate, because tropical, high-humidity climates largely cannot use free cooling. At 33°C and 80% relative humidity, there are effectively no economizer hours in Johor. Every kW of heat rejection runs through mechanical refrigeration, all year.

So a 1.4 in Malaysia is not a design brochure number. It is an operations number: elevated chilled-water temperatures, tight airflow containment, high part-load chiller efficiency, and a plant that stays tuned after handover. A facility designed for 1.4 that operates at 1.55 has a compliance problem, not a marketing problem.

Measuring PUE properly: sub-metering IT load vs facility load

ISO/IEC 30134-2 requires you to meter IT energy separately from total facility energy, continuously, for a full year. Where you place the IT-side meter changes the number:

- Utility intake gives you EDC, the numerator. This is your TNB meter (or meters).
- UPS output is the most common EIT boundary. It excludes UPS losses (correctly, since they are facility overhead) but still includes downstream distribution losses.
- PDU or rack level is the most accurate IT boundary and the most instrumentation-heavy. It also gives you per-hall and per-tenant visibility, which colocation billing needs anyway.

Three practices keep the number defensible:

1. Annualize. A spot reading on a rainy November night flatters you. The guideline requires the annual average, which drags in every hot afternoon and every part-load weekend.
2. Meter continuously, not monthly. Manual monthly reads cannot reconstruct an honest annual ratio, and they hide the drift you need to catch.
3. Reconcile the boundary. Total intake minus IT load should equal your known facility loads within a few percent. If it doesn't, a meter is wrong or a load is unaccounted for.

This is the exact problem a monitoring platform exists to solve: ingest utility, UPS-output and PDU-level meters into one hierarchy, compute the ratio continuously, and keep the year of interval data your auditor will ask for.

The cooling levers that actually move PUE in Malaysia

Cooling dominates the non-IT load in a tropical data centre, so PUE optimization here is mostly chiller-plant optimization. The levers are the same ones we have covered for commercial plants, so briefly:

- Chilled-water temperature reset. Every degree higher on supply temperature cuts compressor lift. Data halls tolerate warmer water than most operators run.
- Fix delta-T. Low delta-T syndrome forces overpumping and starves chillers of load. Our chilled-water delta-T guide covers diagnosis and fixes.
- Track kW/RT, not just PUE. Plant efficiency in kW per refrigeration ton is the leading indicator; PUE is the lagging one.
- Containment and higher rack inlet temperatures. Hot/cold aisle separation lets you raise setpoints without hotspots, which enables the first two levers.

None of these are one-off projects. They are setpoints and operating disciplines that decay without measurement.

WUE and CUE: the other two letters in the guideline

The MIDA guideline does not stop at PUE. It recommends a design WUE of 2.2 m³/MWh or lower (per ISO/IEC 30134-9), with continuous monitoring, disclosure in the annual sustainability report, and improvement to 2.0 m³/MWh within 10 years. New sites should avoid Peninsular areas with a water stress index of 0.8 or higher.

Water is not a theoretical constraint. Johor's roughly 15 operating data centres draw about 18 million litres a day against a projected 384 MLD by 2030, and Ranhill SAJ estimates the state needs around 1,000 MLD of new potable supply by 2030 beyond plants already under construction. Johor has responded by tightening approvals and pushing new facilities toward reclaimed water.

CUE (carbon usage effectiveness, ISO/IEC 30134-8) is calculated for grid-supplied sites as grid emission factor × PUE. Appendix 2 of the guideline gives the Energy Commission's 2021 factors: 0.758 kgCO₂/kWh for Peninsular Malaysia, 0.425 for Sabah, 0.198 for Sarawak. So a Peninsular facility at PUE 1.5 reports a CUE of about 1.14 kgCO₂ per IT kWh, and every PUE improvement flows straight into the carbon report.

The utility squeeze: RP4 tariffs and the 85% utilisation rule

The regulatory pressure is matched by tariff pressure. Under RP4 (1 July 2025 to 31 December 2026), the Peninsular base average tariff rose from 39.96 to 45.62 sen/kWh, and the structure became voltage-based with separate energy, capacity, network and retail charges. Medium- and high-voltage consumers now pay explicit demand-based charges; MV general commercial pays RM89.27/kW capacity plus RM97.06/kW network per month. We break the structure down in our RP4 demand charge explainer.

On top of that, revised TNB Electricity Supply Agreements target phantom demand: data centres must reach 85% utilisation of their contracted demand within four years, with a penalty of RM8.50 per kW of monthly shortfall billed as a line item, plus a Connected Load Charge framework spanning six years from connection.

The context is scale: JLL expects Malaysia's live capacity to more than double to 2,055 MW by end-2026, and the government's own projection has data centres consuming nearly a third of Malaysia's electricity supply by 2035: 73,274 GWh, up from 10,544 GWh in 2026. Regulators tying incentives and tariffs to measured efficiency is the predictable response.

PUE is maintained, not designed: why continuous monitoring decides compliance

A commissioned plant drifts. Condenser tubes foul, sensors go out of calibration, a bypassed valve stays bypassed, IT load grows into (or never grows into) the design assumption. Seasonal weather moves the annual ratio even when nothing is wrong. The gap between design PUE and annualized PUE is where DESAC compliance is won or lost.

That makes the monitoring stack part of the compliance stack:

- Continuous sub-metering at the utility, UPS-output and hall level, so annualized PUE per ISO/IEC 30134-2 is computed from data rather than reconstructed for an audit.
- Anomaly detection on plant efficiency and PUE trend, so a failed sensor or drifting setpoint is caught in days, not discovered in the annual number.
- Regulatory reporting. Large electricity consumers in Malaysia already face EECA (Energy Efficiency and Conservation Act 2024) obligations; PUE, WUE and CUE disclosures sit naturally alongside EECA and ISO 50001 reporting in one system.

CobiNeural does this as an overlay on existing BMS, PLC and SCADA systems, the same platform Malaysian factories and commercial buildings use for energy monitoring, with the metering hierarchy, alerting and EECA/ISO 50001 reporting a data centre needs to keep an annualized PUE defensible.