Lenovo Neptune Immersion Cooling Solution for AI Racks

The Lenovo Neptune Immersion Cooling Solution for AI Racks is a purpose-built single-phase liquid immersion platform designed to address the extreme thermal challenges of next-generation AI training and inference infrastructure. The system submerges entire GPU rack assemblies in a dielectric fluid bath, enabling direct heat transfer from processors, GPUs, memory modules, and power delivery components without the thermal resistance penalties of traditional air or even direct liquid cooling architectures. With per-rack power densities routinely exceeding 100 kW, the Neptune Immersion platform eliminates hot-spot formation and allows continuous, sustained compute operation at full TDP across all installed accelerators simultaneously.

The solution integrates tightly with Lenovo's ThinkSystem server portfolio and is specifically validated for high-density GPU nodes such as those powered by NVIDIA H100, H200, and Blackwell-generation accelerators. The immersion tank is engineered with precision fluid circulation, filtration, and thermal management subsystems that maintain dielectric fluid at optimal operating temperatures, routing heat to a facility-side heat exchanger or a dry cooler loop. This architecture enables data center operators to achieve Power Usage Effectiveness (PUE) values approaching 1.03–1.10 in optimized deployments, representing a dramatic reduction compared to air-cooled equivalents that typically operate at PUE 1.4–1.6. The closed-loop fluid management system includes real-time monitoring of fluid temperature, flow rates, and dielectric properties to ensure long-term fluid integrity and hardware protection.

Launched in Q3 2025, the Lenovo Neptune Immersion Cooling Solution targets hyperscale AI cloud providers, national research computing centers, enterprise AI factories, and co-location operators building dedicated AI zones. The system ships as a modular, rack-scale unit with standardized facility interconnects for cooling water supply and return, enabling deployment in both new construction and retrofit data center environments. Lenovo's end-to-end integration approach—spanning server hardware, immersion tanks, fluid management units, and remote monitoring software—reduces integration complexity and accelerates time-to-production for AI infrastructure projects.

Ideal for

• Hyperscale AI model training clusters requiring sustained full-TDP operation across thousands of GPU accelerators with minimal thermal throttling
• Enterprise AI factories deploying dense GPU racks in facilities with constrained raised-floor space where air cooling is physically insufficient
• National supercomputing centers and research institutions building next-generation GPU clusters that must meet aggressive energy efficiency mandates
• Co-location and edge data center operators establishing dedicated AI zones with high power density in facilities not designed for conventional air cooling at these loads
• Generative AI inference farms requiring continuous low-latency GPU operation with consistent thermal baselines to maintain response time SLAs
• Retrofit deployments upgrading existing data center footprints to support 100 kW+ AI racks without major civil or HVAC infrastructure overhaul

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.