Midea Building Technologies is taking a system-level approach to AI data center cooling as rising rack densities put new pressure on both power delivery and thermal management. At Data Centre World Asia 2026 in Singapore, the company will showcase a power-cooling hyperconverged architecture, a full-stack liquid-cooling portfolio and a Keppel collaboration aimed at demonstrating how high-density AI infrastructure can operate more efficiently.
AI data centers are running into a problem that cannot be solved with faster chips alone: the infrastructure surrounding those chips must deliver more power and remove more heat within increasingly tight physical and energy constraints.
Midea Building Technologies (MBT) is responding by bringing power and cooling closer together at the system level. At Data Centre World Asia 2026 in Singapore, the company is showcasing what it describes as an industry-first power-cooling hyperconverged architecture, alongside a full-stack AI data center cooling portfolio and a strategic collaboration with Keppel.
The move reflects a broader change in data center engineering. AI accelerators are driving much higher rack power densities than conventional CPU-based workloads, putting simultaneous pressure on electrical infrastructure, cooling capacity, water consumption and available floor space.
The International Energy Agency estimates that global data center electricity consumption will roughly double from about 485 TWh in 2025 to 950 TWh by 2030. Electricity consumption from AI-focused data centers is expected to grow even faster, tripling during that period.
That makes cooling efficiency an increasingly important part of AI infrastructure, rather than simply a facilities-management concern.
MBT’s proposed architecture attempts to address the power and thermal sides together. Developed with CLOU Electronics, its AIDC Converged Infrastructure White Paper describes a stack combining electrical power, energy storage, magnetic-bearing cooling sources and coolant distribution units (CDUs).
The concept is significant because conventional data center architectures generally separate power and cooling engineering. As AI workloads become more concentrated, however, the two systems increasingly influence each other. Power availability determines how much compute can be deployed, while cooling capacity determines whether that compute can operate reliably at sustained utilization.
A hyperconverged approach could potentially allow operators to optimize those constraints together.
MBT’s full-stack cooling portfolio provides a more immediate example of that strategy. The company is showcasing a Magnetic CDU that combines a magnetic-bearing cooling source and coolant distribution in one system. MBT says the design can reduce equipment footprint by up to 70% and support a power usage effectiveness (PUE) below 1.2 under suitable operating conditions.
Those are vendor-reported performance figures, not independent benchmarks, and actual results will depend on facility design, workload, climate and operating conditions.
The portfolio also includes an air-cooled magnetic-bearing centrifugal chiller with a coefficient of performance (COP) of up to 5.4 and an industrial-grade CDU rated for 2.6 MW of cooling capacity.
The emphasis on air cooling alongside liquid cooling is notable. Liquid cooling is becoming increasingly important for high-density AI systems because conventional air cooling becomes harder to scale as thermal loads rise. But water availability, local climate and facility design can make different cooling architectures more or less attractive.
That is particularly relevant in Southeast Asia.
The IEA expects electricity consumption from data centers in Southeast Asia to more than double by 2030, with Singapore and southern Malaysia emerging as important regional data center hubs.
For operators in the region, the challenge is not simply obtaining additional electricity. Facilities must also manage heat in hot and humid environments while addressing sustainability requirements and, in some markets, water constraints.
MBT’s collaboration with Keppel illustrates how those considerations can translate into a deployed infrastructure project. The companies are highlighting the Gui’an Midea Cloud Data Centre as an example of combining MBT’s HVAC capabilities with Keppel’s digital infrastructure expertise.
According to MBT, the facility can achieve as much as 7,654 hours of free cooling annually while maintaining a PUE below 1.2. Again, those figures come from the companies and should be treated as project-specific claims rather than a universal performance benchmark.
The broader industry is moving in the same direction: data center infrastructure providers are increasingly competing on how much useful computing capacity can be delivered from a constrained supply of electricity, cooling and physical space.
That changes the definition of efficiency.
For traditional enterprise workloads, improving cooling efficiency might reduce operating costs. For AI infrastructure, better thermal performance can potentially allow operators to deploy more accelerators within the same power envelope or avoid expensive facility expansions.
The IEA’s latest analysis underscores the physical nature of the problem. It says AI server power density increased 11-fold between 2020 and 2025 and could rise another fourfold by 2027. It also estimates that an advanced server rack could have peak power demand equivalent to roughly 65 households by 2027.
Those trends explain why cooling is becoming a strategic component of AI data center design.
The competitive landscape includes established infrastructure companies such as Schneider Electric, Vertiv and Eaton, as well as chip and platform vendors such as NVIDIA, which increasingly influence data center architecture through accelerator, networking and rack-level designs.
MBT’s approach is differentiated less by treating cooling as a standalone product category and more by attempting to connect thermal management with power and storage at the infrastructure level.
Whether that architecture becomes a mainstream model will depend on deployment economics, interoperability, reliability and independently validated performance. But the direction is clear: AI infrastructure is forcing data center engineering disciplines that once operated separately to become more tightly integrated.
As AI compute density continues to rise, the winning data centers may not simply be those with the most powerful accelerators. They may be the facilities capable of supplying and cooling those accelerators most efficiently.
Market Landscape
AI is turning power density and thermal management into fundamental constraints on data center expansion. The IEA expects global data center electricity demand to reach about 950 TWh by 2030, while AI-focused facilities grow substantially faster than the wider market.
That is accelerating investment in direct-to-chip liquid cooling, CDUs, magnetic-bearing chillers, heat recovery, energy storage and integrated power-and-cooling architectures.
Midea’s strategy fits this broader transition toward system-level AI infrastructure. Its full-stack cooling portfolio targets different points in the thermal chain, while the proposed hyperconverged architecture attempts to coordinate power, storage and cooling as a unified infrastructure platform.
The challenge will be proving that these integrated designs can deliver reliable performance at scale while reducing total cost, energy consumption and deployment complexity.
Top Insights
- Rising AI rack densities are forcing data center operators to rethink power delivery and cooling as interconnected infrastructure challenges.
- Midea’s proposed architecture combines electrical power, energy storage and thermal systems into a single AI data center infrastructure concept.
- Liquid cooling and CDUs are becoming increasingly important as conventional air cooling faces higher thermal loads from AI accelerators.
- Southeast Asia’s growing data center market creates a strong test case for efficient cooling in hot, humid and water-constrained environments.
- Future AI data center efficiency may depend on maximizing useful compute from limited power and cooling capacity rather than simply adding capacity.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
