Paid on the Booking: Malaysia’s 5.65 GW Data-Centre Grid Draws 1.26 GW

The Number That Matters Is the Smaller One

Peninsular Malaysia's utility, Tenaga Nasional Berhad (TNB), told analysts on 28 August 2026 that it had 42 data-centre projects in its system at the end of June 2026, carrying 5.65 GW of maximum demand. On the same slide it reported 1,256 MW of actual data-centre load for June 2026¹. The grid has been booked for roughly four and a half times what data centres are drawing from it.

For anyone planning compute in Malaysia, or valuing a Malaysian data-centre platform, the commercial question is who pays for the 4.4 GW gap. TNB management's answer, as relayed by a broker, is that the customer does: "all data centre customers are signed under five-year take-or-pay agreements, allowing TNB to recover its capital expenditure regardless of actual electricity consumption"². If that holds, the idle megawatt is an operator's cost, not the utility's risk. The terms that would let anyone price that cost have not been published.

What Exists, and on What Basis

Every headline figure here comes from TNB, the party that benefits from a large, contracted pipeline. Only one independent number sits alongside them, and it measures something different.

Three distinctions decide how these numbers can be used.

Maximum demand is a ceiling, not a draw. The 5.65 GW is the demand TNB has agreed to be able to serve at the connected sites. The 8.35 GW adds 2.21 GW under construction and 0.49 GW where only the electricity supply agreement (ESA) is signed¹. Neither is consumption, and neither is evidence that the capacity will be used.

The 1,256 MW is measured, but by the seller. TNB labels it "Actual Load Utilisation". The deck does not say whether it is an average or a peak, and no independent meter data exists in the public record. BMI's 925.6 MW of live IT capacity is the only outside reference⁴. It is nameplate IT load, not grid draw, so the two cannot be reconciled directly. What can be said is that the independent figure sits in the same range as TNB's measured load, not its booked demand.

The 12.9 GW is a projection. It comes from the energy ministry, via the government's own economic outlook⁹. It is not a measurement, and it is larger than everything currently contracted.

The Gap, Measured Twice

TNB's quarterly series shows data-centre load of 603 MW in June 2025, then 710, 850, 1,054 and 1,256 MW in June 2026¹. Load roughly doubled in a year, a 108% rise.

Booked demand grew more slowly. CIMB Research, citing end-June figures, put total maximum demand at 5.7 GW in June 2026, up from 3.5 GW a year earlier³, a 63% rise on its own two figures. Taking load over CIMB's maximum demand at each date, the draw rate rose from about 17% (603 MW over 3.5 GW) to about 22% (1,256 MW over 5.7 GW). The numerator is TNB's and the denominator is the broker's. The two denominators agree to within 0.05 GW at June 2026.

So the gap is closing, but slowly: in absolute terms, connected-but-undrawn demand grew from about 2.9 GW to about 4.4 GW over the year.

One reason is visible and one is not. The visible one: TNB's own stage split shows 7 projects with 4.46 GW completed in FY2026 alone, against 35 projects with 1.19 GW completed earlier¹. Newly energised halls take time to fill, so a low draw rate is partly a ramp effect. The invisible one: that split does not reconcile with CIMB's 3.5 GW a year earlier, and a separate parliamentary reply cited by EdgeProp put 36 operating sites at about 4.5 GW of planned supply capacity through Q1 2026¹⁰. Until TNB publishes a cohort-level series, no one outside the utility can say how much of the 4.4 GW gap is ramp and how much is capacity booked ahead of demand that may never arrive.

The utility has disclosed the booking and the draw. It has not disclosed the price of the difference: the take-or-pay floor, the minimum-demand percentage or a RM/kW charge.

What the Idle Megawatt Costs, and What Is Missing

The take-or-pay claim rests on a single relay. It appears in a Business Today report of RHB Research's meeting with TNB management, dated 11 July 2026². It is not in the August analyst deck, which mentions ESAs only as a project stage¹, or in the Energy Commission's tariff statement⁶. CIMB's September coverage does not mention it³. The five-year term should therefore be read as management-reported and unfiled. The charge level, the share of maximum demand it covers, and whether it falls away once a site ramps are all absent.

The energy price is partly documented. The Energy Commission set an average base tariff of 45.40 sen/kWh for July 2025 to December 2027, with customers classed by low, medium and high voltage⁶. The statement names no data-centre or ultra-high-voltage class. The figure most often quoted for data centres, about 60 sen/kWh, comes from analysts who attended a TNB briefing, as reported by EdgeProp⁵. Hong Leong Investment Bank put high-voltage users with low load factors at 58–70 sen/kWh, and Public Investment Bank estimated up to RM63m a year in extra cost for a 100 MW facility⁵. Reuters, citing an industry official and a government official, put the increase at 10–14% before surcharges for major consumers⁷. These are analyst estimates of a tariff, not a published schedule.

The low-load-factor band matters here. A site drawing a fifth of its booked capacity is, by construction, a low-load-factor customer against its contract. If the take-or-pay floor is set on contracted maximum demand, the effective cost per kWh actually used rises as the draw rate falls. That is an inference from the structure, not a disclosed figure. It cannot be quantified until the floor is published.

On the utility side, the de-risking is visible in capex. TNB spent RM7.1bn in the first half of FY2026, of which RM2.3bn went to network investment for demand growth¹. The Budget 2027 speech added that TNB would invest RM15bn in the grid. EdgeProp notes that the speech did not explain how this relates to TNB's existing regulated programme¹⁰. A take-or-pay book is what lets a regulated utility commit that capital ahead of load.

Constraints and Refusals

Concentration is the main physical limit. Johor holds 29 of the 61 secured projects and 67% of secured maximum demand¹, so one state's transmission, land and water carry most of the pipeline. Johor has already narrowed its door. The state reportedly halted new Tier 1 and Tier 2 approvals in 2025. In March 2026 the chief minister said only Tier 3 and Tier 4 air-cooled developments would be approved: "If a data centre demands a high intake of water, we will not approve it"⁸. No counts of refused applications have been published.

No cancelled ESAs or data-centre write-downs have been disclosed by TNB or the operators in these sources. That is an absence of disclosure, not evidence that none have occurred.

What This Establishes, and What It Does Not

Established, on operator primary data: Malaysian data centres had connected about 4.5 times more maximum demand than they drew in June 2026. The draw rate is rising, from about 17% to about 22% in a year. Two-thirds of secured demand sits in Johor.

Not established: the take-or-pay term itself (broker-relayed management statement, five years, no filing); the charge level; the data-centre tariff (analyst estimates around 60 sen/kWh, no published class); whether the reported load is peak or average; and how much of the gap is ramp rather than overbooking. The finding applies to TNB's Peninsular system. It says nothing about Sarawak or Sabah, and nothing about IT utilisation inside the halls.

The Planning Judgement

Treat 1,256 MW, not 5.65 GW or 8.35 GW, as the size of Malaysia's data-centre power market today. Treat the bookings as a liability held by whoever signed them.

For an operator or a tenant passing through power costs, the decision test is simple. Before committing to Malaysian capacity, obtain the ESA's minimum-demand percentage, its charge per kW, and the ramp schedule it allows. Then price the site at a 20–25% draw rate for its first year, because that is what the system as a whole is drawing. For an investor, a platform's contracted megawatts in Johor are an obligation to pay as much as a claim to capacity. Until the take-or-pay terms are filed, the cost of the booked-but-idle megawatt is the largest unpriced line in the Malaysian data-centre case.