Walrus Pump is expanding its data-center liquid-cooling portfolio as rising AI compute density pushes thermal management closer to the center of infrastructure design. The Taiwan-based pump specialist will showcase TPMS, CMP and TPHK-S pump platforms at TAIWAN EXPO USA 2026, targeting in-row and in-rack coolant distribution units (CDUs) and immersion cooling systems.
As AI accelerators consume more power, the challenge facing data-center operators is no longer simply how to install more computing capacity. They also need to remove the resulting heat reliably without consuming excessive space or energy.
That is creating a growing market for direct liquid cooling, coolant distribution units and immersion cooling, and it is putting traditionally specialized components such as pumps into a more strategic role.
Taiwan-based Walrus Pump is targeting that market with three pump platforms designed for different liquid-cooling configurations: the TPMS horizontal centrifugal pump, CMP DC sealless pump and TPHK-S submersible centrifugal pump.
The company will showcase the systems at TAIWAN EXPO USA 2026 from September 24 to 26, positioning the portfolio around AI and high-performance computing (HPC) thermal management.
The TPMS is designed for in-row CDUs, where it uses what Walrus describes as an ultra-slim IE5 permanent-magnet motor. The company says the resulting pump assembly is approximately two-thirds smaller than comparable configurations and incorporates leak-detection sensors for real-time monitoring.
The CMP takes a different approach for in-rack CDUs. Its mechanical configuration can be adapted for 4U or 6U installations, with an adjustable outlet angle intended to simplify integration into space-constrained rack environments. Walrus says its standard 4U configuration provides 20% higher performance than comparable products, a company claim that should be evaluated against the specific operating conditions and competing pump designs.
The third platform, TPHK-S, is designed for immersion cooling. Its return-flow design is intended to reduce coolant leakage and contamination, while different materials for the pump body and shaft seal allow the system to be configured for different liquid applications.
These distinctions highlight an important characteristic of AI cooling infrastructure: there is unlikely to be one universal thermal architecture for every deployment.
High-density GPU servers may use cold plates connected to CDUs, while some specialized systems can use immersion cooling. Facility operators also have to account for rack layout, coolant chemistry, flow requirements, maintenance procedures and the compatibility of cooling infrastructure with different generations of IT hardware.
Uptime Institute’s 2026 cooling survey, based on 1,034 respondents, reflects the industry’s increasing attention to direct liquid cooling. Its research says direct liquid cooling continues to expand, particularly as AI workloads push data centers toward higher densities.
Uptime’s broader 2026 data-center survey also found that average rack densities are continuing to rise, with more operators reporting peak rack densities of 30kW or higher. The organization identifies AI and other high-density computing workloads as major factors behind the trend.
The thermal challenge is becoming more pronounced as AI systems scale. The International Energy Agency says AI server power density increased 11-fold between 2020 and 2025 and could increase another fourfold by 2027.
That trajectory makes pump efficiency and physical integration increasingly relevant. Every watt consumed by cooling infrastructure reduces the efficiency of the overall data-center system, while larger pumps and associated equipment can compete with IT infrastructure for valuable facility space.
Walrus Pump is consequently positioning its technology beyond the sale of individual pumps. The company says it is developing modular configurations and system-integration capabilities that allow pump specifications and supporting components to be matched to a customer’s particular cooling architecture.
That approach could become increasingly important as data-center operators seek liquid-cooling systems that can accommodate multiple generations of AI hardware. Uptime Institute recently highlighted the challenge of creating IT-agnostic direct-liquid-cooling systems capable of supporting different hardware vendors within a shared facility infrastructure.
The company is also developing brushless DC (BLDC) motor technology and preparing for potential future microchannel lid-cooling applications. Those efforts point toward a broader evolution in cooling hardware, where pumps become more digitally controllable, compact and tightly integrated with the thermal-management architecture.
The market is still developing. Uptime Institute notes that direct liquid cooling remains concentrated in high-density environments where air cooling is increasingly impractical, rather than being a universal replacement for conventional cooling.
For AI infrastructure suppliers, however, that high-density segment is already significant. The IEA says data-center electricity demand is accelerating as AI investment expands, while its analysis shows the power density of AI servers rising rapidly.
That makes thermal management a critical part of the AI hardware stack. GPUs and accelerators may determine how much compute a rack can deliver, but pumps, CDUs, heat exchangers and cooling loops determine whether that compute can operate reliably within the facility’s thermal limits.
Walrus Pump’s latest portfolio therefore reflects a broader shift in AI infrastructure: cooling components are becoming system-level technologies. As racks become denser, the ability to control coolant efficiently, reliably and within increasingly constrained physical footprints could become an important differentiator for the next generation of AI data centers.
Market Landscape
AI is driving data centers toward higher rack densities and more sophisticated cooling architectures. Uptime Institute’s 2026 research shows peak rack densities of 30kW or more becoming increasingly common, while its cooling research identifies direct liquid cooling as an expanding technology for high-density environments.
The IEA reports that AI server power density increased 11 times between 2020 and 2025 and is projected to increase another fourfold by 2027. That rapid increase is making thermal management a core infrastructure consideration alongside power delivery and networking.
The market is consequently moving toward a mixture of technologies, including cold-plate liquid cooling, CDUs, immersion cooling, two-phase cooling and more advanced facility-level heat-rejection systems. No single approach currently dominates every AI deployment.
Top Insights
- Walrus Pump is targeting AI and HPC cooling with three pump platforms designed for in-row CDUs, in-rack CDUs and immersion cooling.
- The TPMS targets compact in-row CDU installations, while CMP focuses on space-constrained in-rack cooling architectures.
- TPHK-S is designed for immersion cooling with a return-flow configuration intended to reduce coolant leakage and contamination.
- Rising AI rack density is increasing demand for direct liquid cooling where conventional air cooling becomes increasingly difficult to deploy efficiently.
- Walrus Pump is expanding from component manufacturing toward modular customization, pump configuration and broader cooling-system integration.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
