Power Before Compute: Why Asia-Pacific’s AI Data Centre Boom Is Becoming an Energy and Infrastructure Story

Aug 4

Executive Summary

Asia-Pacific’s artificial-intelligence build-out is usually described as a race for chips, cloud capacity and computing power. Increasingly, however, the decisive questions are more physical: where can a project secure reliable electricity, how quickly can it connect to the grid, what will cooling require, and whether local infrastructure and communities can support the proposed scale. In other words, the region’s data-centre boom is becoming an energy and infrastructure story.

The headline numbers are substantial. Deloitte expects approximately US$800 billion of data-centre investment across Asia-Pacific by 2030. CBRE recorded a regional high of US$11.6 billion in direct data-centre investment during 2025, while JLL projects Asia-Pacific capacity to expand from 32 gigawatts (GW) to 57 GW by 2030. These figures are not directly comparable; one is a forecast, one measures direct investment and one measures capacity, but together they show the scale and breadth of capital now moving towards the sector.

Demand is not the only variable. The International Energy Agency (IEA) estimates that global data-centre electricity consumption rose by 17% in 2025 and could roughly double from 485 terawatt-hours (TWh) in 2025 to 950 TWh in 2030. Electricity use by AI-focused facilities is expected to grow even faster, roughly tripling over the same period. Yet data centres can often be planned and constructed faster than the generation, transmission and grid equipment needed to power them. The result is a widening gap between announced computing ambitions and deliverable infrastructure.

This gap is reordering the Asia-Pacific market. Singapore is releasing new capacity selectively and tying growth more closely to energy efficiency and green-power pathways. Johor has absorbed significant spillover demand but is applying greater scrutiny to power, water and local impact. Thailand is attracting a growing volume of digital-infrastructure applications. India combines strong underlying demand with a large gap between operating capacity and projected AI requirements. Australia is integrating data centres into formal power-system planning as connection requests accelerate.

For developers and capital partners, the implication is clear: land and a compelling demand narrative are no longer enough. A credible data-centre proposition must be supported by a verifiable route to power, a realistic connection date, a water and cooling strategy, appropriate contracting, capable delivery partners and a structure that can withstand delays or changing technology. The opportunity is therefore broader than the data-centre building itself, but so is the diligence required to distinguish a viable project from an ambitious pipeline entry.

From Compute Demand to Infrastructure Demand

AI is changing both the amount and the shape of data-centre demand. Traditional facilities were designed around relatively predictable cloud, storage and enterprise workloads. AI training and inference require high-performance accelerators, denser racks, faster networking and significantly more cooling. CBRE reports that the average size of new data-centre builds in Asia-Pacific now exceeds 100 megawatts (MW), while the IEA expects the peak power demand of a single advanced server rack by 2027 to be comparable with that of 65 households.

This is why additional computing demand becomes infrastructure demand so quickly. A high-density campus requires more than servers and a shell. It needs substations, transformers, switchgear, uninterrupted power systems, backup arrangements, cooling equipment, fibre routes, water infrastructure where applicable, physical security and a grid connection capable of supporting a large and relatively inflexible load. Many of these components have their own permitting processes, supply constraints and construction timetables.

The mismatch in development speeds is becoming material. The IEA notes that technology projects can move quickly while generation, transmission and other energy infrastructure require longer planning periods and substantial upfront investment. In its 2026 update, the agency also identified tighter supply chains for transformers, gas turbines, chips and other critical equipment, alongside planning systems stretched by a rising volume of project applications.

Power availability should therefore not be treated as a utility-procurement item to be solved after site selection. It is a core development input that shapes location, phasing, capital expenditure, operating costs, financing and the eventual customer proposition. A parcel of land can be acquired; a credible path to 100 MW or more of reliable power cannot simply be assumed.

This changes the meaning of “speed to market”. The fastest project is not necessarily the one that secures land or planning consent first. It is the one that can align site control, permits, grid connection, energy supply, equipment procurement, customer commitments and financing into a deliverable sequence. In the current market, speed to power is becoming a more useful measure than speed to construction.

Asia-Pacific’s Changing Data-Centre Map

Asia-Pacific is not one data-centre market. It is a collection of national and subnational markets with different power systems, regulatory regimes, land constraints, fibre connectivity, customer bases and financing conditions. Growth is shifting accordingly.

Singapore remains a leading regional connectivity and cloud hub, but its experience also illustrates the limits of unconstrained expansion. After an earlier pause on new projects, the government’s Green Data Centre Roadmap set out at least 300 MW of additional near-term capacity, with further growth linked to green-energy deployment. A second call for applications launched in December 2025 made at least another 200 MW available, with the possibility of more through innovative green-power pathways. The message is not that Singapore has turned away from data centres; it is that new capacity must justify its use of scarce land and energy.

Johor has been the clearest beneficiary of capacity constraints across the border. Reuters reported in July 2026 that the state had attracted US$35 billion of investment following Singapore’s 2019–2022 moratorium. JLL estimates cited by Reuters indicate that Johor’s planned capacity, including projects under construction and in the pipeline, could reach 7,000 MW, around eight times current levels. That is a powerful growth signal, but it is not the same as 7,000 MW of commissioned, occupied capacity. It also creates an equally significant requirement for generation, grids, water, roads, equipment and skilled delivery teams.

The scrutiny is already increasing. New projects in Johor must explain how they will source electricity, and the state has restricted two especially water-intensive categories of facility. Selangor is vetting projects for energy and water efficiency and requiring local-content commitments in parts of the value chain. This reflects a shift from measuring success by announced capital alone to asking what infrastructure a project consumes and what economic value it leaves behind.

Thailand is also emerging as a serious destination. The Thailand Board of Investment reported US$23.95 billion of digital-industry applications in 2025, largely driven by major data-centre commitments. Applications are not completed investments, and the conversion rate will depend on power, approvals and execution. Even so, the figures demonstrate that regional demand is widening beyond Singapore and Malaysia towards markets that can offer larger sites, supportive policy and a plausible route to utility capacity.

India presents a different scale of opportunity. In July 2026, Brookfield estimated that around 6.5 GW of AI-related data-centre capacity could come online in India over the following five years. Reuters reported that the estimate referred to the wider Indian market, not Brookfield’s own pipeline, and compared with approximately 1.5 GW of installed capacity at the time. The contrast underlines both the strength of the demand thesis and the size of the delivery challenge. Each additional gigawatt requires coordinated land, power, transmission, cooling, equipment and capital, not simply demand from hyperscalers.

Australia shows how quickly data centres are becoming part of formal electricity planning. At the end of the first quarter of 2026, the Australian Energy Market Operator (AEMO) was tracking 11 large projects representing 5.4 GW of maximum demand through the transmission-connection process, with most still at an early stage. AEMO projects data-centre consumption to reach around 12 TWh, or approximately 6% of grid-supplied electricity, by 2030. Crucially, the operator now treats data centres as a distinct category in long-term forecasting because their growth is faster, more concentrated and less predictable than that of traditional large industrial loads.

These examples point to a broader reordering. Established hubs retain advantages in connectivity, customer density and institutional maturity, but power-secure locations are gaining strategic value. The next phase of the market may therefore be distributed across mature hubs, secondary metropolitan areas, industrial corridors and cross-border clusters rather than concentrated only in the region’s traditional Tier I locations.

Power, Water and the New Development Bottlenecks

“Power availability” is often used as if it were one test. In reality, it contains several separate questions: Is sufficient generation available? Can the grid deliver it to the site? Is connection capacity reserved or merely discussed? What upgrades are required, who will fund them, and when will they be completed? What tariffs, curtailment rights and reliability standards apply? Can the project meet customer commitments on renewable energy without confusing contractual certificates with the physical electricity mix?

These distinctions matter because a market can have adequate generation at a national level and still face congestion at the substation or transmission level. It can also have a large renewable pipeline without the grid capacity, storage or contracting frameworks needed to deliver that electricity to a specific campus. According to Reuters, a 2026 Bain and Standard Chartered report estimated an annual shortfall of US$18 billion in Southeast Asian grid investment by 2035. The same report found that almost all surveyed data-centre operators were willing to pay a premium to avoid connection delays.

Renewable procurement is part of the answer, but not a complete substitute for grid planning. The IEA reported that the technology sector accounted for around 40% of corporate renewable power-purchase agreements signed in 2025. Long-term PPAs can support new generation, improve price visibility and help meet customer decarbonisation requirements. Batteries can smooth short-duration fluctuations, provide backup support and potentially offer grid services. Co-location with renewable resources may reduce transmission pressure in selected locations.

However, the physical energy mix remains important. The IEA expects renewables to become the largest source of additional electricity for data centres globally, but it also projects that coal and natural gas will meet a substantial share of near-term demand. In Southeast Asia and India, coal remains an important pillar of the electricity physically consumed by data centres, with renewables projected to overtake it by 2035. A credible sustainability case therefore needs to address both procurement claims and the actual system into which the project is connecting.

Water and cooling are becoming equally consequential. Higher rack densities are accelerating the adoption of liquid cooling, but design choices affect water consumption, power efficiency, capital cost and operational complexity. Malaysia’s sustainable-development guidelines require projects seeking relevant incentives to declare Power Usage Effectiveness (PUE), Carbon Usage Effectiveness and Water Usage Effectiveness (WUE). They recommend that new facilities avoid locations with a water-stress index of 0.8 or above, target a design WUE of 2.2 cubic metres per MWh or lower, and consider reclaimed or reused water.

The community dimension can no longer be separated from technical design. Reuters reported Malaysia’s first community protests against a data-centre complex in Johor in 2026, amid concerns about water pressure, construction impacts and local environmental change. Operators responded with measures including treated wastewater, closed-circuit cooling and renewable-energy arrangements. These examples show that social licence is not a communications exercise undertaken after construction starts. It is a development condition shaped by resource use, local benefits, transparency and the credibility of mitigation plans.

Where the Wider Infrastructure Opportunity Sits

The data-centre building is only one layer of the opportunity. The wider requirement extends across several infrastructure systems, each with different risk, return and development characteristics.

The first layer is power supply: new renewable generation, firming capacity, storage, utility upgrades and, in selected markets, behind-the-meter or private-wire solutions. The second is grid infrastructure: substations, transformers, high-voltage cables, switchgear, transmission links and grid-enhancing technology. The third is the facility layer: high-efficiency cooling, power-management systems, modular equipment, fire protection and resilient backup arrangements. The fourth includes enabling infrastructure such as fibre networks, water and wastewater systems, access roads, logistics and security.

There is also an opportunity in the operating ecosystem. Large campuses require engineering, maintenance, energy management, cybersecurity, specialist construction, compliance and increasingly sophisticated cooling expertise. Governments are paying more attention to local content and skills because a capital-intensive facility does not automatically create broad economic spillovers. Projects able to build credible local supply chains and training pathways may be better positioned to secure political and community support.

At a regional level, the scale of grid investment required goes far beyond data centres. The World Bank estimates that achieving the ASEAN Power Grid vision by 2045 will require approximately US$800 billion in generation and transmission investment. In May 2026, the Asian Development Bank announced plans to support US$70 billion of new energy and digital-infrastructure initiatives across Asia-Pacific by 2035, including stronger grid interconnection and digital connectivity. Data-centre demand is therefore arriving during a much larger cycle of electrification, industrial growth and energy transition.

This creates potential alignment. A well-structured data-centre offtake can help underwrite new generation or storage. A new substation may unlock both a campus and wider industrial development. Cross-border power trade can connect demand centres with lower-carbon resources. Predictable large loads may improve the utilisation of selected energy assets. But alignment should not be presumed: risks need to be allocated between the data-centre developer, energy provider, utility, government and capital partners in a way that is commercially durable.

The opportunity is therefore not simply to “invest in AI infrastructure”. It is to identify which part of the infrastructure stack has a clear customer, an executable route to permits and connection, suitable contracting and a risk profile that matches the capital being deployed.

What Makes a Data-Centre Project Bankable?

Data-centre demand can make a project compelling, but bankability depends on converting that demand into enforceable, deliverable and financeable arrangements. In Asia-Pacific, bank lending remains central to project-level financing, usually supported by long-term contracts. Platform investments, operating-company transactions, private credit, green loans and other structures are also expanding as projects become larger and sponsors seek additional sources of capital.

Several factors now deserve particular attention.

1. A credible customer and revenue case. The strength, term and structure of customer commitments matter more than a general forecast of cloud or AI demand. Diligence should examine whether capacity is pre-leased, reserved, subject to conditions or merely under discussion; the credit quality and concentration of customers; ramp-up assumptions; renewal and termination rights; and who carries the risk if equipment deployment is delayed.

2. A verified route to power. A utility letter, connection application and executed connection agreement represent very different levels of certainty. A bankable project needs clarity on available capacity, connection works, target energisation, upgrade costs, tariffs, curtailment, backup requirements and the consequences of delay. Energy procurement should also be aligned with customer commitments on price, reliability and emissions.

3. Deliverable design and construction. AI-ready facilities require high-density electrical and cooling systems, specialised equipment and experienced contractors. The development plan should reflect realistic lead times for transformers, switchgear, cooling components and grid works, with appropriate contingency, warranties, performance testing and delay protections. A phased campus may reduce initial capital exposure, but only if each phase has coherent access to power and customers.

4. Robust water and cooling assumptions. The chosen cooling architecture must be compatible with local climate, water availability, equipment density and customer requirements. Projects should quantify consumption under realistic loads, not only design conditions, and establish alternative or reclaimed-water sources where appropriate. PUE and WUE targets are useful only when measurement, reporting and accountability are clearly defined.

5. An appropriate ownership and financing structure. Data-centre projects may separate land and buildings from operations, use special-purpose vehicles, or sit within multi-market platforms. Each structure changes the location of construction risk, operating liability, customer concentration and lender security. Drew & Napier has highlighted potential disputes around delays, SPV liabilities, private-credit exposure, regulation and sustainability-linked financing in Southeast Asia. Clear contractual interfaces and step-in rights are therefore essential.

6. Regulatory and social durability. Approvals can tighten as resource pressure grows. A project that meets today’s minimum technical rules but cannot demonstrate local value, responsible resource use and transparent reporting may still encounter delay or opposition. Community impact, local content, workforce development and environmental performance should be incorporated into the investment case, not appended to it.

7. A realistic refinancing or exit pathway. Stabilised data-centre assets can attract infrastructure funds, real-estate capital, strategic operators and platform investors. However, the exit case should not depend on continued yield compression or assume that every development will reach full occupancy. It should reflect lease maturity, customer concentration, remaining capital expenditure, power security, technology adaptability and the depth of buyers for the specific market and structure.

Bankability is ultimately an evidence chain. Demand supports contracts; contracts support revenue visibility; power and permits support delivery; capable sponsors and contractors support execution; and appropriate risk allocation supports financing. Weakness in any link can delay the entire project.

Risks That Could Disrupt the Boom

The first risk is pipeline inflation. Announced investment, promotion applications, planned capacity, connection requests, construction starts, commissioned capacity and occupied capacity describe different stages. Adding them together creates an exaggerated picture of supply. AEMO’s disclosure that most of the 5.4 GW in its Australian transmission-connection process remained early-stage is a useful example of why project status matters.

The second risk is infrastructure delay. The IEA has warned that grid queues and critical-equipment constraints can place a meaningful share of planned projects at risk of delay. JLL reports that average connection waits in primary data-centre markets globally exceed four years. Even where local timelines are shorter, energisation can still depend on upstream transmission, generation and equipment that sit outside the developer’s direct control.

The third risk is cost escalation. Cushman & Wakefield estimates that Asia-Pacific data-centre construction costs increased by an average of 10% during 2025 as demand and supply-chain pressure intensified. Higher specifications for liquid cooling, sustainability and resilience add further cost. Delays can compound this through contractor claims, extended interest carry and lost customer revenue.

The fourth risk is demand and technology uncertainty. AI adoption is expanding rapidly, but future workloads, hardware efficiency and deployment patterns remain uncertain. Training may favour very large centralised campuses, while inference can require more distributed capacity close to users. Efficiency improvements may reduce energy per task even as overall consumption rises. Projects designed around one customer type or one density assumption therefore need enough adaptability to remain relevant.

The fifth risk is energy and environmental mismatch. A project can secure contractual renewable instruments while still connecting to a carbon-intensive or congested grid. Water-efficient design can still face opposition if local supply is perceived as scarce. Sustainability-linked financing creates additional reporting and greenwashing risk if performance definitions are vague or progress cannot be independently measured.

The sixth risk is financial and commercial concentration. Large facilities may depend on a small number of customers, significant upfront borrowing and continued access to refinancing. Currency mismatch, interest costs, customer termination, slower occupancy or sponsor weakness can materially change the economics. Large amounts of capital may remain available for credible projects, but it will not make poorly allocated risk disappear.

Finally, data centres sit at the intersection of technology policy, critical infrastructure and geopolitics. Data-sovereignty rules, chip export controls, cybersecurity requirements, foreign-investment screening and changing incentive regimes can influence where workloads are located and which partners can participate. Cross-border projects need to consider these constraints early, particularly where capital, equipment, operators and customers come from different jurisdictions.

What This Means for Developers and Capital Partners: The ALFA Lens

For developers, the most important change is sequencing. Site origination should begin with a power and infrastructure thesis, not with land alone. The initial data room should contain evidence of grid capacity, connection status, required upgrades, energy-supply options, water conditions, fibre routes, planning constraints and the project’s proposed phasing. A site without this evidence may be a long-dated land position rather than a data-centre project.

For capital partners, market demand should be separated from project readiness. A country can have an excellent long-term digital-growth story while an individual proposal remains unfinanceable. Diligence should classify every headline figure by status: announced, applied, permitted, contracted, under construction, energised or operating, and test how much capital is required to reach the next de-risking milestone.

Cross-sector partnerships should also be formed earlier. Utilities, renewable developers, storage providers, engineering specialists, operators, contractors and potential customers increasingly need to shape the project together. This can create more complexity at origination, but it reduces the risk of discovering late that the facility design, connection plan and customer requirements are incompatible.

Developers should consider where adjacent infrastructure can be integrated and where it should remain separate. A co-located energy asset may improve reliability and support a stronger sustainability case, but it also introduces generation, merchant-price and operational risks that may not belong in the same vehicle as the data-centre real estate. The right answer depends on contracts, counterparties, regulation and the risk appetite of the capital involved.

Local value creation is becoming part of competitive positioning. Projects that can demonstrate efficient resource use, workforce development, local procurement, tax contribution and useful infrastructure upgrades are more likely to retain public support. This is particularly important in markets where communities are beginning to question whether highly capital-intensive facilities provide sufficient local benefit.

Finally, flexibility has value. Phased development, modular systems, adaptable cooling, diversified customers and more than one credible energy pathway can protect a project against shifting workloads and infrastructure delays. Flexibility should not be used to disguise a lack of commitments, but it can prevent a long-lived asset from being locked into assumptions that change before completion.

Conclusion

Asia-Pacific’s data-centre boom is real, but the most important constraint is no longer the availability of a compelling AI narrative. It is the availability of infrastructure that can convert demand into operating capacity.

The region is likely to see continued growth across established hubs and emerging clusters, supported by cloud adoption, AI, data localisation and expanding digital economies. Yet the markets and projects that progress fastest will be those that solve power, grid connection, cooling, water, delivery and community impact as an integrated proposition.

This broadens the opportunity beyond data-centre ownership. Generation, storage, grids, substations, cooling, water systems, fibre and specialist services all sit within the same development cycle. It also raises the standard of execution. Capital will increasingly distinguish between a project that has land and ambition and one that has a credible route to energisation, customers and long-term operation.

In the next phase of Asia-Pacific’s AI build-out, compute may create the demand, but power and infrastructure will determine what is actually built.

Sources:

Report Generated by The ALFA Group

Theo Webb