Malaysia's 300MW AI data centre: what it means for power infrastructure
Malaysia's data centre pipeline picked up another headline this month. For anyone supplying power equipment into that market, the interesting number is not the one in the headline — it is what that number becomes once cooling and conversion losses are added to it.
1. What was signed, and what it commits
Key ASIC Berhad (KLSE: 0143), a Malaysian ASIC design house, and CT Vision (International) Holdings (HKEX: 994), a Hong Kong-listed group working in renewable energy development and construction, signed a ten-year memorandum of understanding on 3 September 2026 to explore a green AI data centre in Malaysia.
The filing is explicit about what an MoU is worth. It records intent and creates no legally binding obligation on either party. Feasibility studies, site selection, detailed engineering and commercial structuring come first, with definitive agreements targeted for the fourth quarter of 2026.
Two figures in the announcement survive into any serious planning: an aggregate IT load of about 300 MW within five years of the commercial operation date, and a design power usage effectiveness of not more than 1.20.
2. What 300 MW of IT load means at PUE 1.20
The 300 MW figure counts computing equipment only. Power usage effectiveness (PUE) is the multiplier that turns it into what the site actually draws, and the project's 1.20 figure is a ceiling rather than a forecast:
300 MW × 1.20 = 360 MW of total facility demand.
That leaves up to 60 MW for cooling, power conversion and distribution losses. At that upper limit the overhead would rank as a large industrial load — and it has to be built, powered and kept running before a single accelerator earns anything.
Renewable supply is part of the plan. The announcement puts green energy — wind, solar, biomass and similar sources — at not less than 50% of consumption. That is a floor for the associated and on-site generation to clear, not a completed power purchase arrangement.
3. What a site this size has to buy
Once IT load reaches this level, the electrical system stops being a line item and becomes the project.
Step-down and distribution transformers, switchgear, UPS systems, battery strings, bypass arrangements and backup generation all scale with the load. Each layer adds its own losses, its own floor space and its own maintenance access requirements.
The UPS layer is where the architecture decision shows up. In an N+1 configuration the margin is one module. In 2N the plant carries two independent paths, each capable of supporting the protected load.
The protected load is not the 300 MW headline. Not every facility load sits behind a UPS at this scale, and how the site splits into power blocks decides what each block carries. Block architecture, redundancy target, load distribution and specified battery runtime then set the final figure — all settled at design stage, and all expensive to revisit once the cable trays are up.
4. AI loads raise the density, not just the total
AI halls pack far more power into a rack than a conventional enterprise room, and that changes the engineering questions in three ways.
Cooling has to follow the heat, which is what pushes PUE targets to 1.20 and below. Distribution has to carry higher current per rack. And the UPS plant has to be designed around actual and peak load behaviour rather than nameplate totals alone — our breakdown of rated, actual and peak power covers that distinction in detail.
Redundancy is not optional at this level. Maintenance has to happen without dropping load, which is what the bypass and distribution design is really for. The 10 kVA parallel installation in our project notes is a small-scale version of the same argument: one module out for service, the load stays up.
5. Where the medium-power work actually sits
A 300 MW campus is hyperscale territory. Its UPS plant will be built from the largest three-phase systems available, or from distributed and rack-level alternatives, specified by the operator's own engineering team against a global supplier list.
That is not where most of the region's buying happens. The projects that grow up around a hyperscale anchor — regional data centres, telecom facilities, disaster recovery sites, enterprise server rooms, and the industrial loads sharing their grid connection — are where mid-range three-phase work lands.
TAFENG supplies three-phase online UPS systems, isolation transformers and voltage stabilizers under the BLAZING POWER brand for this middle segment — applications with configurable input and output voltages, defined backup times and critical loads to protect. Choosing between them starts from the load and the redundancy target rather than the kVA label, which is what the sizing guide for hospitals, banks and factories walks through.
Utility voltage comes first. Southeast Asian sites do not share one standard, and the input configuration has to match the local supply before capacity is even discussed.
6. Read the scale before the headline
The 300 MW figure is a market signal rather than a confirmed build.
The MoU is non-binding. Feasibility studies, site selection, detailed engineering, commercial structuring and definitive agreements — targeted for the fourth quarter of 2026 — all have to follow before any construction decision exists.
For power-equipment suppliers the signal is directional rather than commercial. Southeast Asian AI capacity is increasingly planned alongside dedicated generation, tighter efficiency targets and larger electrical systems. As computing capacity grows, the transformers, UPS plant, distribution equipment and backup power behind it have to be specified at the same time — the same pattern shows up in transformer demand across Africa and North America.
Source: CT Vision (International) Holdings voluntary announcement, HKEX, 4 September 2026; Key ASIC Berhad bourse filing, 4 September 2026; The Edge Malaysia, 4 September 2026.
FAQ
Is the 300 MW Malaysia data centre confirmed?
No. It is the target IT load in a memorandum of understanding that creates no legally binding obligation. Feasibility studies, site selection and definitive agreements — targeted for the fourth quarter of 2026 — come before any construction decision.
What does PUE 1.20 mean for a 300 MW data centre?
PUE 1.20 is a ceiling rather than a forecast. At that figure, 300 MW of IT load corresponds to 360 MW of total facility demand, leaving up to 60 MW for cooling, power conversion and distribution. That overhead is a large industrial load in its own right, which is why efficiency targets are a capital expenditure question rather than an environmental one.
How much UPS capacity does a 300 MW data centre need?
No single UPS capacity follows from the 300 MW IT figure. The protected load comes first, then the power-block design and the redundancy architecture. N+1 adds spare capacity within each block; 2N provides two independent paths, each capable of supporting the protected load. Battery runtime and load distribution shape the final system from there.
Is the Malaysia AI data centre running on 100% renewable energy?
The announcement sets a floor of not less than 50% from green sources including wind, solar and biomass. Higher percentages have circulated in secondary coverage; the filing states the 50% floor and describes renewable supply as part of the development plan.
What power infrastructure does an AI data centre need beyond UPS?
Step-down and distribution transformers, switchgear, battery strings, bypass arrangements and backup generation, all sized against the total facility demand rather than the IT load. Lead times on the transformer and switchgear layers are worth confirming early in the programme.
