Midea’s Clivet brand has unveiled a next-generation magnetic-bearing centrifugal compressor and liquid-cooling solution designed for large-scale AI data centers, where rising computing density is increasing demand for efficient, reliable thermal management. Presented at the 2026 OCP Global Summit, the technology combines independently controlled compressor shafts, flexible single- and two-stage operation, and cooling-mode transition controls. Midea says the design can reduce compressor power by up to 40% at the same cooling capacity and increase maximum inlet flow by approximately 70% under low-pressure-ratio conditions. Those performance figures are company-reported and have not been independently verified in the supplied announcement.
Midea Targets AI Data Center Cooling Challenges With New Compressor Design
The rapid expansion of AI infrastructure is creating new demands for data center cooling. As facilities scale toward gigawatt-level power capacity, operators must remove heat from increasingly dense computing systems while managing energy consumption, water availability, equipment footprint, and the risk of cooling interruptions.
At the 2026 Open Compute Project (OCP) Global Summit, Midea’s Clivet brand presented a new approach to this challenge: a magnetic-bearing centrifugal compressor designed to operate across a wider range of cooling conditions, alongside a full-chain liquid-cooling solution. Dr. Yunxiao Ding, director of the Research Institute at Midea Building Technologies, presented the architecture at the event’s GW-Scale Open AIDC forum. Clivet said the presentation received the forum’s 2026 GW-Scale Open AIDC Excellent Technical Presentation Award.
The technology reflects a broader shift in AI data center infrastructure, where cooling is increasingly treated as an integrated system-design problem rather than a standalone equipment purchase.
Why AI data centers need more flexible cooling
High-density AI servers generate substantial heat, making cooling capacity and operational continuity critical to data center design. Clivet’s presentation describes a scenario in which cold-plate supply temperatures rise toward 45°C, expanding opportunities for free cooling—the use of ambient conditions to reduce reliance on mechanical refrigeration.
Dry heat rejection can also help address water constraints. However, prioritizing free cooling while retaining mechanical cooling as a backup creates engineering challenges. Compressors must perform efficiently under changing pressure ratios, and the system must transition between operating modes without disrupting heat removal.
Traditional compressor designs can face trade-offs between high-capacity performance in hot conditions and efficiency at lower pressure ratios in milder weather. Switching between free cooling and mechanical cooling introduces another risk: rapid pressure changes can destabilize the system and affect cooling delivery. Meanwhile, data center operators need more performance from equipment without expanding its physical footprint.
Independent drives enable flexible compressor operation
Clivet’s design uses independent variable-frequency drives for two compressor shafts, allowing each impeller to operate at a different speed. It also supports selectable single-stage and two-stage compression.
Under high-pressure-ratio conditions, both stages can work together to deliver compression performance. When pressure ratios are lower, the system can switch to single-stage operation by shutting down and bypassing the second stage. This configuration is intended to avoid unnecessary power consumption and improve flow performance in conditions where conventional two-stage operation may be less efficient.
According to Midea, the compressor can reduce power consumption by up to 40% at the same cooling capacity, while increasing maximum inlet flow by approximately 70% under low-pressure-ratio conditions. The announcement does not provide independent test results, a detailed baseline for those comparisons, or a complete account of the operating conditions behind the figures.
The practical value will depend on how the compressor performs across real-world weather conditions, changing server loads, and the full cooling system—not just in an individual operating mode.
Transition control focuses on cooling continuity
A second element of the design addresses the switch between free cooling and mechanical cooling. Clivet says its system uses single-stage startup and preconditioning to manage these transitions.
When mechanical cooling is needed after free cooling, the single-stage path is intended to establish system flow and cooling capacity more quickly. In the opposite direction, preconditioning aims to reduce fluctuations as the system returns to free cooling.
These controls are important because cooling interruptions can put high-density computing workloads at risk. For AI infrastructure operators, reliability depends not only on nominal cooling capacity but also on how equipment responds to power interruptions, changing environmental conditions, and mode switching.
Clivet says its approach can eliminate cooling interruptions during transitions. That claim will need validation under specific deployment conditions, including system-level testing and long-duration operation.
Compact design and wider ecosystem integration
The compressor architecture also incorporates outboard bearings into recesses in the volute, alongside multifunctional components and compact motors. Midea says these measures preserve the axial length of a traditional single-shaft unit while creating room for heat exchange and maintenance.
Midea is also positioning itself within the broader AI infrastructure ecosystem. The company said it appeared on the NVIDIA MGX ecosystem’s participating-company list at the summit. That participation indicates ecosystem involvement, but does not by itself establish a formal product certification, deployment agreement, or endorsement.
For operators planning gigawatt-scale AI facilities, the next test is whether integrated cooling architectures can deliver measurable improvements in energy use, water management, serviceability, and reliability at scale. Clivet’s announcement outlines a design aimed at those priorities, but deployment-level performance data will be needed to assess its practical impact.
As AI infrastructure expands, compressor efficiency remains only one part of the equation. Cooling continuity, integration with liquid-cooling systems, and predictable operation across changing conditions will increasingly determine whether thermal-management platforms can support the next generation of compute capacity.
Market Landscape
AI data center cooling is evolving alongside higher rack power densities and growing demand for liquid cooling. Cold-plate systems, coolant distribution infrastructure, chillers, compressors, and heat-rejection equipment must work together to maintain stable operating conditions.
Clivet’s announcement targets three connected requirements: improving compressor efficiency across varying operating conditions, maintaining cooling continuity during transitions, and reducing equipment footprint. Its free-cooling-first approach also reflects operator interest in reducing mechanical refrigeration demand where ambient conditions permit.
Competition in this market extends beyond individual cooling components to system integration, operational reliability, energy performance, and compatibility with emerging AI infrastructure architectures. Midea’s participation in the NVIDIA MGX ecosystem is relevant industry context, but the announcement does not establish a specific joint product deployment.
The key question for buyers is whether the reported performance gains translate into lower total energy consumption and dependable operation across a complete data center cooling system.
Top Insights
- Midea’s Clivet brand unveiled a magnetic-bearing centrifugal compressor and liquid-cooling solution designed for next-generation, GW-scale AI data centers.
- Midea reports up to 40% lower compressor power at equivalent cooling capacity, though independent testing and detailed comparison conditions were not supplied.
- Independent variable-speed drives enable selectable single- and two-stage operation, targeting performance across different pressure ratios and cooling demands.
- Transition controls aim to maintain cooling continuity when systems switch between free cooling and mechanical cooling.
- Compact mechanical integration and ecosystem participation are part of Midea’s strategy to address larger AI infrastructure requirements.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
