Malaysia's Data Centre Gate Is a Water Margin
Malaysia will not approve data centres in states without spare water. The real test is a reserve margin percentage, and it changes the maths per site.

On 28 October, in the Budget 2027 winding-up debate, two ministers described the same policy from opposite ends of the pipeline. Investment, Trade and Industry Minister Datuk Seri Johari Abdul Ghani told the Dewan Rakyat that a state without sufficient water will not be allowed to host data centres, even if it has the land. Deputy Prime Minister Datuk Seri Fadillah Yusof, in a written reply as Energy Transition and Water Transformation Minister, supplied the mechanism: projects in areas with low water reserve margins will not be considered, and developers must hold at least two days of consumption on site.
Read together, they are one gate — and the gate is not a moratorium. It is a percentage.
That distinction matters more than the headline. A moratorium is a date you can wait out; a reserve-margin test follows the site and moves as demand grows. For anyone holding land options in a data centre corridor, the question changed from "is there water?" to "how much spare capacity does this distribution system have, and what fraction of it does my hall eat?"
What was actually decided
There is no blanket freeze. Projects will keep being approved, but assessed on water availability, electricity supply and the investor's financial strength — MIDA with state governments on water, Tenaga Nasional on power.
The pipeline, per Johari's breakdown: 117 approved data centres, 55 operating, 19 under construction, 16 approved but not started, and 27 at various stages with power expected from off-grid arrangements. At full load, w.media reports him putting electricity demand at 23–23.5 GW.
Two weeks earlier, MITI gave Parliament a different snapshot: 57 operating centres drawing 6,237 MW, with 60 more needing 11,172 MW. The counts drift because approval, energisation and commissioning are different events; neither version changes the policy.
Three conditions now sit on top of the old approvals:
- Water. Location screened against state reserve margin and the state water operator's recommendation; two days of on-site storage; alternative sources (recycled, reclaimed, rainwater) expected.
- Power. No household-style subsidy; market rates; and payment for the full contracted capacity even when actual draw is lower. Johari's example: contract 300 MW, use 100 MW, pay for 300 MW.
- Local content. A 30% requirement on data centre-related development, monitored through six-monthly implementation reports, alongside access to compute for local universities and firms.
Add Fadillah's renewable condition — at least 30% of energy capacity or consumption from renewables, with CRESS for direct green supply and SELCO for on-site solar — and the approval file stops looking like a property deal and starts looking like a utility contract.
Reserve margin is a percentage, so the same hall passes or fails by postcode
The part that gets lost between Putrajaya and a site plan: "sufficient water" is not a national quantity but a ratio owned by a treatment and distribution system, and Malaysia's systems are nowhere near equal.
SPAN's 2023 annual report put the national average at 15.3%, but the spread is brutal: Terengganu 27.7%, Perak 27.5%, Pulau Pinang 25.6%, FT Labuan 22.5%, Negeri Sembilan 17.1%, Johor 16.9%, Selangor 13.4%, Pahang 11.0%, Melaka 6.1%, Perlis 4.0%, Kelantan 3.5%, Kedah negative 9.8%. Those are 2023 figures and some states have moved since — Johor was near 14.4% in June 2026 — but the shape holds. The northern states and Melaka were already the wrong postcode for a water-hungry hall before this policy existed.
Now put a number on the hall. Malaysia's sustainable data centre guideline recommends a design water usage effectiveness of 2.2 cubic metres per MWh of IT energy or lower, improving toward 2.0 within ten years, and says new projects should avoid areas with a water stress index of 0.8 or above. A 100 MW facility at full load therefore needs about 5.28 million litres a day at the design ceiling. The ASEAN+3 Macroeconomic Research Office used a lower figure — roughly 4.2 million litres a day per 100 MW, about the daily use of 10,000 people — which is the same physics with a different load factor and cooling design.
Both are defensible, which is why borrowing one is a mistake. The figure that counts is computed from the design: IT load, utilisation, cooling type, and whether the site can run on reclaimed water.
Two days of storage is an emergency buffer, not a drought hedge
The two-day on-site storage rule is easy to misread as drought resilience. It buffers the failures that hit Malaysian water systems: a treatment plant trip, a burst main, a scheduled shutdown, a pump failure. For a 100 MW hall at 5.28 MLD, two days is about 10,560 cubic metres of tankage — a real capital item and footprint, and no protection against a dry spell measured in weeks.
The policy's own framing gives this away. Fadillah described operators' emergency plans escalating through tanker deliveries, static tanks, inter-state transfers and finally rationing. Storage buys time to execute that ladder. It does not remove the hall from it.
The quieter sentence is more consequential: locations will be screened on "recommendations from state water operators." A utility already nursing a thin margin and a non-revenue-water problem has every incentive to say no, and the federal task force has just made that recommendation load-bearing. That is the de facto veto, held by the party with the least interest in your project.
Do the arithmetic before you option the land
You do not need a consultant to run the first pass. You need three numbers from the water operator — production, reserve margin, and whether reclaimed supply is available — plus your own design figures. This is the screening model, standard library only:
"""Screen a proposed data centre against a state's water reserve margin.
Inputs you must get in writing from the state water operator:
- production_mld : treated water the system puts out today
- reserve_pct : design capacity minus current demand, as a % of production
The hall's own demand comes from its design, not from an industry average:
demand_mld = it_load_mw * 24 * load_factor * wue / 1000
(wue in m3 per MWh of IT energy; Malaysia's guideline ceiling is 2.2)
"""
from dataclasses import dataclass
TARIFF_RM_PER_M3 = 5.31 # data centre category, Peninsular Malaysia
STORAGE_DAYS = 2 # on-site buffer required by PETRA
MAX_SHARE_OF_HEADROOM = 0.10 # house rule: one project, max 1/10 of spare
@dataclass(frozen=True)
class Hall:
name: str
it_load_mw: float
wue: float # m3/MWh
load_factor: float = 1.0
@dataclass(frozen=True)
class State:
name: str
production_mld: float
reserve_pct: float
def demand_mld(hall: Hall) -> float:
return hall.it_load_mw * 24 * hall.load_factor * hall.wue / 1000
def screen(hall: Hall, state: State) -> dict:
need = demand_mld(hall)
headroom = state.production_mld * state.reserve_pct / 100
share = need / headroom if headroom > 0 else float("inf")
return {
"hall": hall.name,
"state": state.name,
"need_mld": round(need, 2),
"headroom_mld": round(headroom, 2),
"share_of_headroom_pct": round(share * 100, 1),
"storage_m3": round(need * 1000 * STORAGE_DAYS),
"water_cost_rm_year": round(need * 1000 * 365 * TARIFF_RM_PER_M3),
"verdict": "PASS" if share <= MAX_SHARE_OF_HEADROOM else "FAIL",
}
Run it against a large system with a healthy margin and a small one with a thin margin, and the postcode problem becomes obvious:
100MW evaporative, WUE 2.2, full load | large system, healthy margin | 5.28 MLD | headroom 144.0 MLD | 3.7% | tank 10560 m3 | RM 10,233,432/yr | PASS
100MW evaporative, WUE 2.2, full load | small system, thin margin | 5.28 MLD | headroom 4.0 MLD | 132.0% | tank 10560 m3 | RM 10,233,432/yr | FAIL
100MW evaporative, WUE 2.2, 85% load | large system, healthy margin | 4.49 MLD | headroom 144.0 MLD | 3.1% | tank 8976 m3 | RM 8,698,417/yr | PASS
100MW evaporative, WUE 2.2, 85% load | small system, thin margin | 4.49 MLD | headroom 4.0 MLD | 112.2% | tank 8976 m3 | RM 8,698,417/yr | FAIL
100MW evaporative, WUE 2.0, full load | large system, healthy margin | 4.8 MLD | headroom 144.0 MLD | 3.3% | tank 9600 m3 | RM 9,303,120/yr | PASS
100MW evaporative, WUE 2.0, full load | small system, thin margin | 4.8 MLD | headroom 4.0 MLD | 120.0% | tank 9600 m3 | RM 9,303,120/yr | FAIL
100MW closed-loop, WUE 0.3 | large system, healthy margin | 0.72 MLD | headroom 144.0 MLD | 0.5% | tank 1440 m3 | RM 1,395,468/yr | PASS
100MW closed-loop, WUE 0.3 | small system, thin margin | 0.72 MLD | headroom 4.0 MLD | 18.0% | tank 1440 m3 | RM 1,395,468/yr | FAIL
Three things fall out. A modest WUE improvement buys about 9% less water at full load — helpful, not decisive. Utilisation matters almost as much as design: an 85% load factor cuts demand 15%, which is why committed capacity is a water question too. And even a closed-loop hall at 0.72 MLD fails a 100 MLD system with a 4% margin, because the constraint is spare system capacity, not hall efficiency. Efficiency gets you into the conversation; the margin decides it. The 10% share rule is my own engineering convention, not a regulation — but some threshold like it is what a utility applies when ranking applicants.
The cost column matters less than it looks. At the data centre tariff of RM5.30–5.38 per cubic metre, water for a 100 MW evaporative hall runs about RM10.2 million a year — small beside the power bill, enormous beside a state's tolerance for industrial draw in a dry month. Water is cheap and scarce at once, and only one of those facts reaches a spreadsheet.
Three contracts now matter more than the land title
If you are scoping a project, the approval risk has moved. Land and fibre are no longer the long poles and capital is not scarce: SoftBank's US$50 billion Malaysian intention was announced six days before the ministers spoke.
A water letter, not a water estimate. Get production, reserve margin and the operator's written position on reclaimed supply before committing.
A power contract sized to what you will actually draw. Take-or-pay turns an over-ambitious capacity request into a recurring cost. Johor already illustrates the tension: about 825 MW operational, another 3.9 GW expected in three to five years, including a gigawatt campus in Sedenak, and one analysis projecting roughly 30 TWh a year by 2031 — 18% of Peninsular Malaysia's supply. Grid connection, not generation, is the named bottleneck.
A local content plan with evidence. Thirty percent is a monitored commitment with six-monthly reporting, not a procurement slogan. Consultancy, contracting, engineering and legal work all count, and a plan assembled after commissioning will not survive review.
Since February 2026, approvals for new non-AI facilities have been restricted, so for plain colocation the water gate is not the only gate.
What to watch
Three signals will show how strict this gets.
- Which states say no, and why. The policy names no state. Watch the first rejection and its stated ground — reserve margin, water stress index, or the operator's recommendation; that precedent will matter more than the announcement.
- Whether "financial strength" becomes a bond. A reserve-margin test is cheap to administer; a financial test is not. A deposit or capacity bond would price smaller developers out before they reach the water queue.
- Whether the 30% renewable condition is measured on capacity or consumption. Those are very different numbers at a low load factor, and the wording allows either.
Malaysia is not turning away data centres. It is turning a growth story into a site-by-site utility negotiation. For developers the lesson is unglamorous: water arithmetic is a pre-feasibility item now, and it belongs in the same spreadsheet as the land option.
Sources
- The Edge Malaysia: States without sufficient water supply cannot host data centres, says Johari Ghani
- The Vibes: Data centres face water and power curbs to protect public supply
- The Star: 12.3GW energy capacity approved for data centres as of August, Parliament told
- Free Malaysia Today: Govt tightens conditions for data centres
- w.media: No data center moratorium for Malaysia but states with insufficient water can't host data centers
- The Star: Johor has enough water reserves to meet current demand, says Fadillah
- MIDA: Guideline for Sustainable Development of Data Centre
- AMRO: Malaysia Annual Consultation Report 2026 — The Data Center Boom in Malaysia
- SPAN: Annual Report 2023
- Air Selangor: Water Tariff Review in Peninsular Malaysia and Federal Territory of Labuan
- Free Malaysia Today: Johor's data centre boom faces power supply crunch
- Amanz: Kerajaan Mengetatkan Kelulusan Pembukaan Pusat Data Berdasarkan Tahap Bekalan Air Negeri
author
Editor & Operator
MrDodgerX (MrDodgerX) is an enterprise geospatial systems engineer in Malaysia and the editor behind hejes.my. He runs the briefing pipeline, curates and edits every draft, verifies each factual claim against primary sources, and presses the publish button — nothing ships without his review. He writes about AI, security and Malaysia's tech scene from a builder's perspective. Professional profile and code: the links below.
sources
fact-check verified · 30 OCT 2026
Source details and access dates
- States without sufficient water supply cannot host data centres, says Johari GhaniThe Edge Malaysia · accessed 30 OCT 2026
- Data centres face water and power curbs to protect public supplyThe Vibes · accessed 30 OCT 2026
- 12.3GW energy capacity approved for data centres as of August, Parliament toldThe Star · accessed 30 OCT 2026
- Govt tightens conditions for data centresFree Malaysia Today · accessed 30 OCT 2026
- No data center moratorium for Malaysia but states with insufficient water can't host data centers: Ministerw.media · accessed 30 OCT 2026
- Johor has enough water reserves to meet current demand, says FadillahThe Star · accessed 30 OCT 2026
- Guideline for Sustainable Development of Data CentreMalaysian Investment Development Authority · accessed 30 OCT 2026
- Malaysia Annual Consultation Report 2026 — The Data Center Boom in MalaysiaASEAN+3 Macroeconomic Research Office · accessed 30 OCT 2026
- SPAN Annual Report 2023Suruhanjaya Perkhidmatan Air Negara · accessed 30 OCT 2026
- Water Tariff Review in Peninsular Malaysia and Federal Territory of LabuanAir Selangor · accessed 30 OCT 2026
- Johor's data centre boom faces power supply crunchFree Malaysia Today · accessed 30 OCT 2026
- Kerajaan Mengetatkan Kelulusan Pembukaan Pusat Data Berdasarkan Tahap Bekalan Air NegeriAmanz · accessed 30 OCT 2026
related

The 30-Day Form Behind Malaysia's AI Seizures
Customs seized RM49.5M of AI servers at KLIA. The offence was a missing 30-day form, and the directive's real gate is a Category 4 entry few read.

Your Phone May Scream on 19 November. No Proof You Heard It
Malaysia's alert test will be judged on cells reached, not handsets. The tender behind it says so — and its message-ID table decides whether your phone rings.

Malaysia's Courier Boom Is a Margin Problem
Five million parcels a day sounds like growth. Malaysia's own data shows volume up 29% and courier headcount down 12% — and the margin question nobody asked.
join the feed
one fresh insight per week. no spam, ever.
weekly digest. unsubscribe anytime.