According to a research report from CITIC Securities, liquid cooling thermal dissipation demand is now expanding beyond GPUs and CPUs to encompass entire cabinet heat sources such as power supplies and optical modules, driving continued industry growth. The AI cabinet liquid cooling investment theme has shifted from single-chip thermal management to full-cabinet thermal control.
With Rubin and Google's TPU V8 entering full-scale mass production cycles, the liquid cooling industry is poised for a concentrated period of order and earnings delivery. CITIC Securities recommends exploring liquid cooling supply chain opportunities along four key investment lines: 1) Air-cooling vendors transitioning to liquid cooling, where existing customer relationships and system integration capabilities represent their greatest assets; 2) Connector manufacturers extending into liquid cooling connectors, which offer strong customer stickiness after entering the BOM and increasing value per generation; 3) Thermal management companies pivoting to liquid cooling, where materials and component expertise extend to system-level solutions, resulting in significant value jumps; 4) Other manufacturers entering liquid cooling from adjacent industries, offering the highest growth elasticity.
Rubin Paradigm: From Hybrid Air-Liquid to Fanless Design, Cabinet Value Reaches New Heights
NVIDIA officially defines Rubin as the world's first AI infrastructure achieving 100% liquid cooling, featuring no fans within the system. It employs 45°C water inlet with single-phase direct-to-chip liquid cooling, substantially reducing mechanical refrigeration requirements and enabling a chiller-free natural cooling path. The VR200 NVL72 cabinet power consumption rises to 190-230kW.
Based on NVIDIA's CES, GTC, and earnings call disclosures, Vera-Rubin launches in January 2026, enters production ramp-up by the end of May, begins mass production in Q3, starts deliveries in Q4, and scales significantly through late 2026 into 2027. CITIC Securities anticipates the production timeline will be accelerated.
Beyond NVIDIA: Full Liquid Cooling Represents an Industry-Wide Architecture Shift
Google's TPU v7 (Ironwood) features 9,216 liquid-cooled chips per pod with 42.5 ExaFLOPS of cluster FP8 compute and approximately 980W TDP per chip. 100% liquid cooling is a mandatory requirement, with third-generation liquid cooling infrastructure already exceeding 1GW in production. TPU v8 adopts fourth-generation liquid cooling technology.
According to the AAI conference, AMD's Helios / MI455X integrates 72 GPUs using liquid cooling manifolds with quick disconnects for direct liquid cooling, with shipments beginning at the end of Q3 2026 and volume ramping in Q4. Schneider Electric's reference design supports 246kW racks scalable to 10.4MW. This industry-wide architectural shift means liquid cooling demand now extends from single-chip solutions to full-cabinet end-to-end integration.
Overseas Infrastructure Gap Creates Smooth Export Channels for Chinese Manufacturers
According to TrendForce data, liquid cooling penetration in global AI data centers is projected to rise from 14% in 2024 to 33% in 2025, reaching 40% by 2026. In terms of regional market structure, North America represents the largest volume while Southeast Asia shows the fastest growth—precisely the regions with the tightest domestic production capacity and highest receptivity to Chinese supply chains.
Overseas leaders are seeing surging order growth, with the Middle East and Southeast Asia emerging as new growth frontiers. Chinese manufacturers are validating order execution with smooth overseas expansion progress.
National Standards Rollout Drives Mass Production as Gross Margins Poised for Recovery
2026 marks the concentrated year for liquid cooling standard implementation, pushing the industry from project-specific customization toward standardized mass production, resolving interface incompatibility and specification variance issues. The liquid cooling equipment sector is expected to continue its downward trend through H1 2026, with gross margin recovery anticipated in H2 as economies of scale materialize. Competitive focus shifts from supply capability to system integration proficiency.
Risk Factors
Demand shortfall risk; intensifying competition pressuring profitability; Rubin mass production and delivery delays below expectations; technology roadmap evolution risk; overseas geopolitical and trade uncertainty risks.