NVIDIA's (NVDA) fiscal 2026 fourth quarter and full-year results significantly surpassed market expectations, solidifying the high-growth trajectory of the liquid cooling sector. The deep integration of liquid cooling in the GB300 and the mandate for 100% liquid cooling in the next-generation Rubin platform mark a substantial increase in power density per rack, elevating liquid cooling from an optional feature to a standard component of AI computing infrastructure. Furthermore, NVIDIA has delegated procurement authority for core components like cold plates and CDUs to its partners. Under the pressure of technological iteration in liquid cooling, this open trend is expected to persist, creating a historic opportunity for domestic manufacturers with core technologies such as micro-channel cold plates and high-density CDUs to qualify for NVIDIA's supplier list.
The architectural upgrade of the Rubin platform's liquid cooling system is driving a significant increase in value per system. Compared to the GB300's hybrid cooling, Rubin achieves 100% full liquid cooling, an upgrade to micro-channel cold plates, and modular, cable-and-hose-free connections. This reduces maintenance difficulty and leakage risk while increasing the value of liquid cooling components per rack, potentially raising the overall value benchmark for the industry. The technical barriers of micro-channel technology are expected to create opportunities for domestic manufacturers to benefit from both volume and price increases.
NVIDIA's supply chain is becoming more open, presenting a golden entry window for domestic liquid cooling companies. The shift from the H100's "designated sole supplier" model to the GB300's "whitelist + ODM autonomous procurement" approach, where NVIDIA delegates procurement of key components, signifies this change. The next-generation Vera Rubin platform's significantly higher power density per rack, coupled with its "100%" liquid cooling design that supports 45°C water cooling without traditional chillers, places higher demands on technologies like micro-channel cold plates and high-density CDUs. This open trend is anticipated to continue, offering a historic chance for qualified domestic players.
The rapid adoption of ASICs is emerging as a second major growth driver for the liquid cooling market. GPUs and ASICs are complementary, each excelling in different scenarios. Cloud giants like Google, Amazon, and Microsoft are rapidly scaling their self-developed ASICs. With chips like the TPUv7 reaching 980W per chip and Trainium3 supporting liquid cooling for 144 chips, power consumption is exceeding the limits of air cooling. ASICs' market share is projected to rise to 27.8% by 2026, driving rapid growth in liquid cooling demand.
Domestic energy efficiency policies are being strengthened and implemented, accelerating the adoption of liquid cooling. Under the "Dual Carbon" strategy, major policies in cities like Beijing and Shanghai mandate that data centers with a PUE above 1.35 will face tiered electricity pricing, and inefficient centers may be phased out. As one of the most core and effective technologies for reducing PUE, liquid cooling has become essential for data center compliance, avoiding penalties, and achieving green transformation, with policy benefits now fully materializing.
Investment recommendations are driven by three factors: rising computing infrastructure demand, NVIDIA's outperformance and high liquid cooling penetration, tightening energy efficiency policies, and ASIC growth as a secondary driver. It is advised to focus on three core areas: domestic component manufacturers with leading technology that could qualify for the GB300/Rubin whitelist; liquid cooling solution providers deeply adapted for high-power ASIC chips; and industry leaders with full-stack delivery capabilities and high order visibility. Companies to watch include those specializing in cold plates, CDUs, and liquid cooling pumps and valves.
Risks include potential shortfalls in downstream computing demand, delays in technology R&D and iteration, failure to meet NVIDIA's supply chain qualifications or secure orders, and slower-than-expected adoption of ASICs and their corresponding liquid cooling solutions.
Comments