APAC is not facing a uniform, present-day 42% data-center shortage. The figure is the upper end of a CBRE forecast published in 2025, which estimated that the region could face a 15–25 gigawatt (GW) capacity shortfall by 2028—equivalent to a projected 25%–42% gap against demand, depending on the scenario.
The more important conclusion is narrower and more durable: APAC is adding data centers rapidly, but power-secure, high-density, AI-ready capacity is arriving more slowly than demand. That creates opportunities for infrastructure developers and utilities, but also execution, valuation and reliability risks for investors, cloud buyers and businesses planning AI deployments.
As an Amazon Associate I earn from qualifying purchases.
The short version
- AI, cloud adoption and digital services are increasing demand for data-center capacity across APAC.
- CBRE estimated in May 2025 that the region could be short 15–25 GW by 2028.
- The widely repeated “42%” figure represents the upper end of a forecast range, not a current shortage in every APAC country.
- The binding constraint is increasingly electricity, grid access, cooling and high-density design—not simply the amount of physical floor space.
- A large announced project pipeline does not guarantee operational, commissioned or customer-ready capacity.
CBRE’s May 2026 APAC outlook said the structural pressure had not disappeared. It identified AI, cloud adoption and “neocloud” providers as major demand drivers, while noting that power availability remained a major bottleneck.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
For businesses and investors, the practical question is therefore not “How many megawatts are being announced?” It is “How much suitable capacity can actually be energized, cooled, connected and delivered in the required market and timeframe?”
#1 Best Overall
- Save valuable floor space: 6U wall mount server cabinet Dimensions: 13.78" H x21.65" W x17.72" D.Maximum mounting depth is 14.2"
- Keep critical network equipment secure: glass door and side panels are lockable to prevent unauthorized access. Front door can be installed on either side of the front of the cabinet to satisfy your door swing orientation preference
- Easy equipment configuration: Fully adjustable mounting rails and numbered U positions, with square holes for easy equipment mounting with top and bottom punch-out panels for easy cable access
- Durability: Made of high quality cold rolled steel holds up to 110lb (50kg) (Easy Assembly Required)
- PCI & HIPPA and EIA/ECA-310-E compliant
What the 42% estimate actually means
CBRE’s 2025 research reported APAC installed data-center supply of approximately 13.3 GW and expected expansion to about 30 GW over the following two to three years. Against projected demand, the research indicated a possible 15–25 GW shortfall by 2028.
The 42% number came from the upper end of the report’s estimated 25%–42% gap against forecast demand. It should not be rewritten as “APAC is currently 42% short of data centers.” Nor does it mean every country or city will experience the same deficit.
The forecast depends on how supply and demand are defined, which facilities are counted, and whether capacity is operational, contracted, under construction or merely planned. As of 2026, the available CBRE update confirms continuing growth and power constraints, but it does not establish that the exact 42% scenario has already materialized.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesThat distinction matters financially. A forecast shortfall can support strong demand for power-secure sites and infrastructure, but it does not guarantee that every proposed campus will be built, occupied or profitable.
“Supply” is more complicated than a pipeline number
Data-center capacity can refer to several different stages of delivery:
| Category | What it means | Why it matters |
|---|---|---|
| Operational IT capacity | Servers can run now and customers can use the capacity. | The most relevant measure for immediate availability. |
| Built or shell capacity | The building exists, but power systems, equipment or commissioning may be incomplete. | A completed structure may still be unusable. |
| Powered land | The site has a credible utility connection and contracted capacity. | Often more valuable than a finished shell with uncertain energization. |
| Under construction | Physical development is underway but the facility is not yet available. | Delivery remains exposed to equipment, financing and permitting delays. |
| Planned pipeline | A project has been announced or proposed. | It may still lack financing, permits, customers or grid access. |
| AI-ready capacity | The facility supports dense GPU clusters, suitable cooling, electrical distribution and high-speed networking. | Generic data-center megawatts are not automatically equivalent to AI-ready megawatts. |
| Latency-appropriate capacity | The facility is close enough to users, datasets or regulated workloads. | A remote power-rich site may not replace a constrained metropolitan location. |
Cushman & Wakefield reported a 19.4 GW APAC development pipeline in 2025, including 3.7 GW under construction and 15.7 GW planned. That is evidence of substantial investment and expected demand—not proof that 19.4 GW is immediately available.
Why AI changes the data-center equation
Traditional enterprise and web workloads generally spread computing demand across many lower-density racks. AI training and some inference deployments concentrate substantially more computing power in fewer locations.
CBRE said AI-focused facilities can require more than twice the power density per rack of traditional data centers. Coverage of the 2025 research cited AI rack densities above 40 kilowatts, compared with roughly 5–15 kW for many legacy environments. Those figures are illustrative ranges, not universal specifications for every AI deployment.
AI-ready facilities typically need:
- Higher rack power density: GPU systems can draw several times the power of conventional server racks.
- Advanced cooling: Direct-to-chip liquid cooling, coolant distribution units or other solutions may be needed when air cooling is insufficient.
- Stronger electrical systems: Transformers, switchgear, busways, backup generation and power distribution must support concentrated loads.
- High-speed interconnects: Training clusters need low-latency, high-bandwidth communication between GPUs.
- Appropriate floor loading: Dense servers and cooling equipment can impose requirements older facilities were not designed to meet.
- Large contiguous footprints: Training workloads often operate more efficiently in campus-scale deployments than in scattered small rooms.
- Reliable data and network access: AI workloads require substantial storage, ingestion and east-west traffic between systems.
This is why a conventional 10 MW facility cannot automatically be treated as equivalent to a 10 MW AI facility. The electrical, thermal, networking and physical design may be unsuitable even when the headline IT load appears identical.
Power is the critical path
Delivering AI infrastructure involves a chain of dependencies:
Customer demand → data-center reservation → land and site selection → utility study → generation and interconnection → transmission and distribution upgrades → construction → equipment delivery → commissioning.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →A project can be delayed at any stage. In many APAC markets, constructing the building is not the hardest part. The more difficult step may be securing a utility connection with sufficient capacity on a reliable timetable.
Common constraints include:
- Utility interconnection queues.
- Insufficient transmission capacity.
- Substation and transformer shortages.
- Generation that has not yet been built or contracted.
- Local restrictions on electricity use.
- Water availability and environmental approvals.
- Community opposition.
- Long lead times for electrical and cooling equipment.
Deloitte estimated that total APAC electricity demand could rise almost 50% between 2024 and 2035. Data centers are therefore competing with industrial expansion, transport electrification, household demand and other economic priorities. They are not operating in an empty power system.
CBRE’s 2026 update said average new facilities now exceed 100 MW and that regional data-center electricity consumption had nearly doubled from 2020 to 2024, with further substantial growth expected. These projects can be economically attractive, but their size makes grid planning, energy procurement and community acceptance more consequential.
Rank #2
- Save valuable floor space: 12U wall mount server cabinet Dimensions: 24.25" H x21.65" W x17.72" D. MAXIMUM MOUNTING DEPTH is 14.2".
- Keep critical network equipment secure: glass door and side panels are lockable to prevent unauthorized access; Front door can be installed on either side of the front of the cabinet to satisfy your door swing orientation preference
- Easy equipment configuration: Fully adjustable mounting rails and numbered U positions, with square holes for easy equipment mounting with top and bottom punchout panels for easy cable access
- Durability: Made of high quality cold rolled steel holds up to 110lb (50kg) (Easy Assembly Required)
- PCI & HIPPA and EIA/ECA-310-E compliant
Where APAC’s pressure points differ
APAC is not one market. Power, land, connectivity, regulation, labor and customer demand vary widely by country and city.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Singapore
Singapore offers mature connectivity, a strong technology ecosystem and dense customer demand. Its constraints are equally clear: limited land and power availability. Expansion has increasingly spilled into nearby markets, particularly Johor in Malaysia and other parts of Southeast Asia. Singapore remains valuable for latency-sensitive and connectivity-heavy workloads, but additional capacity is difficult and expensive to deliver.
Malaysia and Johor
Johor benefits from proximity to Singapore and can offer larger sites for hyperscale and regional cloud campuses. That does not remove execution risk. Developers still need to verify grid capacity, water availability, connectivity, permitting and local policy rather than assuming proximity to Singapore guarantees a suitable site.
Indonesia
Indonesia has expanding digital demand and can serve domestic and regional workloads. Site selection depends heavily on power quality, land, connectivity, regulation and the ability to deliver reliable service at the required scale.
India
India’s large domestic digital economy supports strong cloud and data-center demand. Mumbai, Chennai, Bengaluru and other markets have different land, power, connectivity and latency profiles. Chennai’s submarine-cable connectivity can be relevant for international traffic, while the best location for domestic inference may differ from the best location for large-scale training.
Free tools Windows power users keep installed
One-click scans. No signup required.
Australia
Sydney and Melbourne remain important hyperscale and enterprise markets. Power availability and development timing are becoming more decisive, while Melbourne may offer greater industrial land availability for some large developments. Australia also illustrates the trade-off between strong demand, high operating costs and the need for credible power delivery.
Japan and South Korea
Both markets have advanced technology ecosystems and substantial enterprise demand. Land, grid conditions, construction economics and access to suitable powered sites can constrain expansion. Korea’s limited development land makes competition for power-secure locations particularly important.
Mainland China
China has large domestic cloud and AI demand, but its market conditions differ from those in Singapore, India or Southeast Asia because of domestic providers, policy, regional power availability and existing supply. Western China may attract additional capacity where power is more available, although latency and network economics can limit the suitability of those sites for some workloads. Regional vacancy and supply conditions can also differ sharply within the country.
CBRE’s APAC market research emphasizes that these markets should be assessed individually rather than ranked using a single “next hub” narrative.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWhy secondary markets may win—and what they sacrifice
Developers are looking beyond traditional hubs because secondary locations may offer:
- Larger and more affordable land parcels.
- More available power.
- Less congested permitting.
- Room to design campuses for high-density workloads from the beginning.
- Lower construction costs or faster delivery.
- Government incentives.
The trade-offs can include higher latency, fewer carrier and cloud interconnection choices, smaller technical labor pools, less mature supply chains and greater dependence on a limited number of utilities or telecom providers.
The right location decision is not simply “cheap land versus expensive land.” It is a comparison of:
- Time to energization.
- Usable power quality and redundancy.
- Connectivity and fiber-route diversity.
- Latency to users and data.
- Water and cooling resources.
- Regulatory and data-sovereignty requirements.
- Local labor and maintenance capability.
- Disaster recovery and physical-security conditions.
Training can sometimes move to a power-rich region because it is less sensitive to end-user latency. Inference often cannot. Inference may need to remain close to customers, factories, financial markets, regulated datasets or real-time applications. Fine-tuning and retrieval workloads sit between those extremes.
Recommended Free Tools
Independent forecasts show why the issue matters
The estimates differ because they measure different things, but they point in the same direction:
Rank #3
- Sturdy:4u server rack is construct from cold rolled steel, with a weight capacity of 110lbs(50kg); Electrostatic powder coat prevents rust and corrosion,quality finish
- Direct use:Open and use, not having to assemble it.Network rack can be placed flat or mounted on the wall,also can be installed vertically under the table
- Design Features:maximum mounting depth of 14 in,cables can be fixed on the side panel;Open frame server rack achieves effortless inspection, replacement and assemble
- Installation:wall mount network rack is easy to install,with instructions or videos for reference;Equipped with multiple accessories, suitable for different needs
- Application:EIA/ECA-310-E Compliant;wall mounted 4u rack fits all 19" racks and cabinets to hold various IT, network, and AV equipment;wall mount rack available in 4U, 6U, and 8U to choose
- CBRE’s 2025 research estimated a potential 15–25 GW APAC shortfall by 2028.
- PwC projected APAC AI-related data-center capacity could rise from 2.2 GW in 2024 to 4.8 GW in 2028, a 21% compound annual growth rate.
- McKinsey’s global scenario projected AI-related data-center demand could increase from about 44 GW to 155 GW. That is global context, not an APAC forecast.
- CBRE reported US$11.6 billion in direct APAC data-center investment during 2025, demonstrating the scale of capital competing for sites and projects.
Growth forecasts are not guarantees. Efficiency improvements, model optimization, higher utilization and changes in AI adoption can reduce demand per task. But cheaper computation can also stimulate more usage, offsetting part of those gains.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the infrastructure gap means financially
For investors and infrastructure companies, scarcity can support pricing power and demand for powered land, substations, cooling systems, grid equipment, fiber and specialist construction. It can also increase the value of projects with contracted customers and credible delivery dates.
However, high demand does not eliminate execution risk. A project can be impaired by:
- Delayed grid connections.
- Higher financing or construction costs.
- Transformer, switchgear or cooling-equipment shortages.
- Customer concentration.
- Changes in AI hardware economics.
- Water restrictions or community opposition.
- Uncertain energy prices.
- Permitting or data-localization changes.
Investors should distinguish between an announced campus, a financed project, a project with secured power and a commissioned facility with paying customers. Those stages carry different risk and should not be valued as though they were interchangeable.
How buyers should evaluate “AI-ready” capacity
A cloud customer, enterprise or AI developer should ask for evidence rather than rely on the label.
Capacity and power
- What is the contracted critical IT load?
- When will the capacity be energized and customer-ready?
- Has the utility connection been approved, contracted and scheduled?
- Are dual feeds and redundancy available?
- What backup generation and fuel arrangements exist?
- Can the site expand in phases?
- Is renewable energy physically delivered, contractually procured or represented through certificates?
AI technical readiness
- What rack densities are supported today?
- Is cooling air-based, direct-to-chip liquid, immersion or hybrid?
- Are coolant distribution and water-treatment systems available?
- Can the floor support the required equipment weight?
- What busway and power-distribution architecture is installed?
- Which GPU servers and cluster topologies are supported?
- What high-speed fabric and storage throughput are available?
- How are commissioning, burn-in and hardware replacement handled?
Connectivity and compliance
- Is the facility carrier-neutral?
- Which cloud on-ramps and submarine-cable routes are available?
- Are fiber routes physically diverse?
- What is the latency to users, data sources and other regions?
- What are cross-connect and egress charges?
- Are data-residency, security and industry compliance requirements met?
Commercial terms
- What minimum commitment and contract term apply?
- Who bears power-price changes or energy pass-through costs?
- Are expansion rights contractual?
- What happens if the promised power date slips?
- Are remote-hands, installation, replacement and support fees separate?
- Can the customer exit if the facility cannot support the promised density?
The relevant comparison is total delivered cost, not just an advertised GPU-hour price. Buyers should include compute, storage, bandwidth, egress, interconnects, orchestration, reserved-capacity premiums, power and cooling charges, minimum commitments and support costs.
What governments and utilities need to solve
Meeting AI demand requires more than approving buildings. Policymakers and utilities need coordinated planning for:
- Transmission, distribution and substation upgrades.
- Faster but credible permitting.
- Large concentrated loads in grid forecasts.
- Firm and lower-carbon power procurement.
- Water, cooling and environmental impacts.
- Fiber and international connectivity.
- Incentives tied to delivered capacity, jobs or infrastructure—not only announcements.
- Regional coordination where workloads, power and network routes cross borders.
Renewable procurement can help with emissions and long-term supply, but it does not automatically guarantee local reliability. Physical power, contractual energy, renewable certificates, firmed supply and grid capacity are different things.
Could efficiency gains eliminate the shortage?
They could reduce the pressure, but there is no basis for assuming they will remove it entirely. More efficient models, quantization, batching, improved utilization and specialized hardware can reduce the resources required for a given task. Distributed inference can also move workloads closer to users.
At the same time, lower computing costs can encourage more companies to deploy AI and more users to generate AI workloads. New power generation and secondary-market campuses may add supply, but those projects still face grid, permitting, cooling, connectivity and workforce constraints.
A regional shortage can also coexist with local oversupply. One city or country may have vacant conventional capacity while another lacks dense, connected, latency-appropriate AI facilities.
Bottom line
The “42%” headline is best understood as a warning about the upper end of a CBRE forecast range for 2028—not as a present, uniform APAC deficit. The underlying issue is real: AI demand is growing faster than the region can reliably deliver power-secure, high-density and connected capacity.
The winners will not necessarily be the projects with the most impressive announced megawatts. They will be the organizations that can secure power, cooling, connectivity, permits and delivery certainty in the locations their workloads actually require.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
