KUALA LUMPUR, 5 AUG 2026 — Malaysia has put a number on what the data-centre build is going to cost its grid. In a written parliamentary reply, Energy Transition and Water Transformation Minister Datuk Seri Fadillah Yusof said the sector is projected to take 31 per cent of the country's electricity by 2035, or 73,274 gigawatt-hours, against 7 per cent and 10,544 GWh today.

That is the whole argument in one line. On 3 August we reported that Thailand has begun requiring proof that the power exists before it will approve a data centre. Malaysia's figures show what the regulators in the region are looking at when they do that.

The system grows; the sector grows much faster

The percentages tell most of the story; the gigawatt-hours behind them tell the rest. Peak demand in Peninsular Malaysia is projected to rise from 21.3 GW in 2026 to 33.5 GW in 2035, a compound annual growth rate of 5.1 per cent. The figures come from the Electricity Supply and Tariff Planning and Implementation Committee, known as JPPPET.

Computed by RECATOOLS5 August 2026
20262035 (projected)
Data-centre electricity use10,544 GWh73,274 GWh
Share of national consumption7%31%
Peninsular peak demand21.3 GW33.5 GW
Implied total consumptionabout 151,000 GWhabout 236,000 GWh

Data-centre figures, shares and peak demand are the minister's, sourced to JPPPET projections. The implied total consumption row is RECATOOLS arithmetic — dividing each data-centre figure by its stated share — and is shown because it is what makes the two percentages legible. Peak demand rising from 21.3 GW to 33.5 GW is an increase of about 57%, also our arithmetic.

Divide each data-centre figure by its stated share and the shape becomes clear. The whole system is expected to grow by roughly half over the period. The data-centre load inside it grows by roughly seven times. The data-centre share of the grid moves from one-fourteenth to nearly one-third not because other demand shrinks, but because this one category is expanding at a rate the rest of the system cannot match.

A 57 per cent rise in peak demand over nine years is a large but manageable planning problem for a utility. It becomes a different problem when a single customer class multiplies sevenfold inside that growth, because that class does not behave like diversified household and industrial load. It is flat, it runs at night, and it does not fall away in a recession. Household and industrial demand peaks and troughs, which is what lets a grid serve more connected capacity than it could ever supply at once. A load that never troughs removes that cushion.

What Malaysia is doing about it

The ministry says the government screens new data-centre applications and expansions through a Data Centre Task Force, and that approvals go only to projects matching local grid capacity planning without compromising the safety and security of electricity supply.

Thailand has written this instinct into its investment rules, and Johor applied it administratively when it began turning projects away. The regional pattern is no longer three governments improvising separately. It is a common recognition that grid capacity, not land or tax, is the binding constraint on where this build goes.

What Malaysia has published, and Thailand has not, is the demand curve the screening is meant to manage. A gate is easier to defend when the number behind it is public.

Singapore is doing the same sum from the other end

The contrast across the causeway is instructive, because Singapore has been living with this constraint for longer and has arrived at a more explicit answer.

Having spent years with new data-centre capacity effectively frozen, Singapore reopened it through capacity allocation calls rather than ordinary approvals. The second of these, DC-CFA2, was launched on 1 December 2025 by the Economic Development Board and the Infocomm Media Development Authority. It allocates around 200 megawatts, requires that half the energy be green, and sets a power usage effectiveness target of 1.25 at full load.

This is a different instrument from a screening committee. Singapore is not assessing applications against available capacity; it is deciding in advance how much capacity exists, publishing the number, and then running a competition for it on efficiency and clean-supply terms. The scarcity is the policy.

Malaysia has the opposite problem and therefore the opposite instrument. It has land, it has generation headroom that Singapore does not, and it has become the destination for the capacity Singapore cannot host — much of it in Johor, immediately across the strait. Screening is the tool you reach for when the constraint is real but not yet binding. Allocation is what you reach for when it is.

The 2035 projection marks when Malaysia expects to cross from screening to allocation.

The part the projection does not settle

A projection is not a plan. This one leaves the expensive questions open: how the additional generation gets built, and by whom. It does not say what share is expected to be renewable, which matters because data-centre operators increasingly arrive with corporate commitments to clean supply and will look for a mechanism to meet them. And it does not say who pays for the network reinforcement that a 12-gigawatt increase in peak demand implies.

That last one is the question the region keeps circling. Thailand has answered part of it with a bank guarantee per megawatt of reserved capacity and a proposed premium tariff. Malaysia's reply describes screening but no equivalent pricing signal. Screening decides which projects proceed; it does not decide who carries the cost of the capacity they consume once they do.

There is a political economy to this that the projection also implies. The Johor-Singapore Special Economic Zone was set up on precisely the complementarity involved: Singapore has the demand and the customers, Johor has the land and the power. An arrangement built on Malaysia supplying what its neighbour cannot only works while Malaysia has the headroom to supply it. The 2035 figure is, among other things, a statement about how long that lasts.

Why the 2035 number is the useful one

Ten-year electricity projections are not forecasts in any strong sense. They are the planning assumption a utility and a regulator agree to work from, and they get revised. The value of this one is not that 31 per cent will turn out to be right.

The government has now committed in writing to a scenario where data centres become the single largest driver of electricity demand growth. Every subsequent argument — about tariffs, about who funds transmission, about whether a project gets approved — now has an agreed baseline to be argued against. Before the reply, an operator and a regulator could each hold a different picture of the decade. They can no longer do that without one of them contradicting a published parliamentary answer.

For anyone siting capacity in the region, the practical reading is simple. Malaysia is not closing the door — the Data Centre Task Force language is about matching capacity, not refusing it. But the era in which a project's power requirement was somebody else's problem, resolved after approval, is closing across ASEAN at roughly the same time in three different jurisdictions. The number Malaysia has just published is the reason.