AI servers are pushing data-center rack densities into territory that conventional air-cooling architectures were not designed to handle. LONGWELL is responding with an AI-era data center cooling package built around EC fan technology, targeting lower cooling energy consumption, faster equipment development and greater resilience for high-density compute environments.
The AI infrastructure boom is creating a problem that has little to do with model performance: heat.
As accelerators become more powerful and server racks become denser, data-center operators are being forced to rethink how air moves through cooling systems. Rack densities that once sat around 15–20 kW are increasingly reaching 60–100 kW, according to LONGWELL, with next-generation AI platforms moving beyond 100 kW per rack.
That puts cooling efficiency directly into the conversation about AI infrastructure economics.
LONGWELL (Ningbo Longwell Electric Technology Co., Ltd.), an EC fan and motor manufacturer founded in 1990, has introduced what it calls its AI-Era Data Center Cooling Pack, centered on the company’s LWBE3G EC plug-fan platform.
The company’s proposition is straightforward: make the fan stage of precision cooling systems more efficient while giving operators greater control and redundancy.
The LWBE3G platform is designed for CRAH and CRAC systems, where arrays of electronically commutated (EC) plug fans can replace or modernize conventional fan configurations. LONGWELL says its FanWall architecture can reduce CRAH fan energy consumption by 38%.
That number matters because cooling is one of the largest non-IT energy loads inside a data center. As compute density increases, simply adding more cooling capacity can increase a facility’s electricity requirements unless the cooling infrastructure becomes correspondingly more efficient.
LONGWELL says a deployment developed with a top-three global precision-cooling OEM delivered several measurable improvements. According to the company, 12 engineering samples were delivered within 35 days, with design and production validation completed successfully on the first attempt. The development cycle from specification lock to mass production was approximately 90 days, compared with an incumbent European supplier’s quoted 14-month development period followed by six months of ramp-up.
The reported operating results included a 38% reduction in fan energy, a 6.5 dB(A) reduction in noise and a reduction in the CRAH system’s PUE contribution from 1.42 to 1.28. LONGWELL also reports zero field failures for the deployment.
Those figures are company-reported rather than independently verified, but they illustrate the kind of optimization data-center operators are increasingly looking for.
The commercial scale of the deployment is also notable. The customer’s first purchase order covered 1,500 units, while 2025 deliveries exceeded 80,000 units under a 2025–2027 framework agreement carrying a 60,000-unit annual minimum, according to LONGWELL.
The company’s broader cooling package extends beyond the server room.
The LWBE3G EC plug fans target CRAH and CRAC applications, while the LWAE3G EC axial fan is designed for condensers and cooling towers. Both sit within an architecture that supports Modbus control and N+1 automatic failover.
That last capability is important in AI facilities, where cooling interruptions can have consequences far beyond temporary discomfort. High-density GPU clusters generate substantial heat continuously, and cooling-system failures can trigger performance throttling, hardware protection mechanisms or even downtime.
LONGWELL is also offering optional bearing-vibration sensors designed to provide 30–90 days of warning before expected end-of-life conditions. Predictive maintenance can be particularly valuable in large data centers because maintenance teams need to service thousands of mechanical components without unnecessarily taking equipment offline.
The development highlights a larger transformation in data-center cooling.
For decades, air cooling was largely optimized around relatively predictable thermal loads. AI computing introduces much higher and more variable heat densities. NVIDIA GPU platforms, for example, have pushed data-center designers toward increasingly dense rack configurations, while hyperscalers and colocation providers are exploring liquid cooling for workloads that exceed the practical limits of traditional air systems.
That does not make air cooling obsolete.
Instead, it creates a more differentiated market. Air cooling remains useful for many workloads and can continue to handle substantial heat loads when paired with efficient airflow management, higher-performance fans and optimized facility design. Liquid cooling becomes more attractive as rack power moves further beyond what air-based systems can efficiently dissipate.
Fan technology therefore remains relevant even in a liquid-cooled future. Air movement is still required elsewhere in the thermal-management chain, including heat rejection, cooling towers and hybrid architectures.
This is where LONGWELL’s strategy becomes broader than a single EC fan.
The company describes its platform as covering the full airflow chain, from internal CRAH and CRAC equipment to external condensers and cooling towers. That gives data-center cooling OEMs a common technology platform across multiple parts of the thermal infrastructure.
The move also reflects the growing convergence between mechanical equipment and software-defined infrastructure.
Modbus connectivity enables fans to participate in facility-management and control systems, while automated failover and condition monitoring introduce capabilities traditionally associated with IT infrastructure into mechanical cooling equipment.
For data-center operators, the ultimate metric is not simply fan efficiency. It is whether the entire thermal system can deliver the required cooling while minimizing energy use, maintaining availability and supporting increasingly dense compute.
LONGWELL’s CEO Bruce Li argues that above 60 kW per rack, the fan stage becomes a determining factor in whether a data center can reach its power-usage-effectiveness targets.
That claim is deliberately strong, but the underlying trend is difficult to ignore.
AI infrastructure is forcing operators to optimize every layer of the physical stack—from GPUs and networking to power delivery and cooling. As rack densities climb beyond 100 kW, thermal engineering is becoming a constraint on how much compute a facility can deploy in a given footprint.
The next generation of data centers will therefore be defined not only by how many GPUs they can install, but by how efficiently they can keep those GPUs operating.
Market Landscape
The data-center cooling market is undergoing a structural shift as AI workloads push rack densities beyond the levels traditionally associated with enterprise computing.
Three approaches are increasingly competing or working together:
- Advanced air cooling: Higher-efficiency EC fans, optimized airflow, containment and improved heat exchangers.
- Direct-to-chip liquid cooling: Liquid removes heat directly from processors and accelerators, making it suitable for higher-density racks.
- Hybrid thermal architectures: Air cooling remains responsible for parts of the facility while liquid systems handle the highest-density components.
Companies such as Vertiv, Schneider Electric, STULZ, Delta Electronics and others are investing in technologies for high-density AI facilities, while GPU platforms from NVIDIA are influencing the thermal requirements of next-generation deployments.
The market opportunity is expanding alongside AI infrastructure investment. The International Energy Agency has warned that electricity consumption from data centers is set to rise sharply through the end of the decade, with AI a major driver of that growth. Cooling efficiency will consequently become increasingly important to both operating costs and grid impact.
For enterprise and hyperscale operators, the decision is no longer simply which cooling technology has the highest theoretical efficiency. Reliability, deployment time, serviceability, controls integration, redundancy and compatibility with existing facilities are becoming equally important.
That gives efficient EC fan systems a role even as liquid cooling gains momentum.
Top Insights
- LONGWELL’s LWBE3G platform targets AI-era cooling, using EC plug-fan arrays to reduce CRAH energy consumption while supporting high-density computing environments.
- The company reports a 38% fan-energy reduction and lower system PUE contribution, potentially helping operators control the rising energy cost of AI cooling.
- Modbus control and N+1 failover allow cooling equipment to integrate more closely with data-center management systems while maintaining resilience during component failures.
- Predictive bearing monitoring can provide up to 90 days of warning, giving data-center maintenance teams more time to plan interventions before fan failures.
- AI rack densities are changing thermal design, creating demand for advanced air cooling, liquid cooling and hybrid architectures across next-generation data centers.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
