CITIC SEC: Liquid cooling shifts from an "optional configuration" to "rigid infrastructure," from "spatial expectations" to "order acceleration"

CITIC SEC released a research report stating that liquid cooling demand is spreading from GPUs and CPUs to whole-rack heat sources including power supplies and optical modules, and the industry continues to expand; the main investment theme for AI rack liquid cooling is shifting from single-chip heat dissipation to whole-rack thermal management. As Rubin and Google TPU V8 enter full mass-production cycles, the liquid cooling industry is about to enter a period of concentrated order and earnings realization. We recommend exploring investment opportunities in the liquid cooling supply chain along four main lines: 1) Air-cooling manufacturers transitioning to liquid cooling, with existing customers and system integration being their greatest assets. 2) Connector manufacturers extending into liquid cooling connectors, with strong stickiness after entering the BOM and value rising with each generation. 3) Thermal management manufacturers transitioning to liquid cooling, extending material/component capabilities to the system level, with a jump in value. 4) Other manufacturing companies making cross-sector forays into liquid cooling, offering the greatest elasticity. CITIC SEC's main views are as follows: Rubin Paradigm: From "Hybrid Air-Liquid" to "Zero Fans," Single-Rack Value Rises Another Notch NVIDIA officially defines Rubin as "the world's first AI infrastructure to achieve 100% liquid cooling," with no fans anywhere in the system, using 45C inlet water and single-phase direct-to-chip liquid cooling, greatly reducing mechanical cooling demand and enabling a "chiller-free" natural cooling path; VR200 NVL72 single-rack power rises to 190230kW. According to NVIDIA's CES, GTC, and earnings calls, Vera-Rubin was released in January 2026, enters production ramp at the end of May, mass production in Q3, deliveries from Q4, and enters a large-scale volume ramp from late 2026 to 2027. We expect the mass-production pace to be brought forward. Beyond NVIDIA, Full Liquid Cooling Is an Industry-Level Architecture Shift Google TPU v7 (Ironwood) has 9,216 liquid-cooled chips per Pod, cluster FP8 compute of 42.5 ExaFLOPS, and single-chip TDP of about 980W, with 100% liquid cooling as a mandatory standard; the third-generation liquid cooling infrastructure already has more than 1GW in production; TPU v8 adopts fourth-generation liquid cooling. According to the AAI conference, AMD Helios / MI455X integrates 72 GPUs, adopts liquid cooling manifold + quick connectors for direct liquid cooling, begins shipping at the end of 26Q3 and ramps in Q4, with Schneider's reference design at 246kW per rack and scalable to 10.4MW. An industry-level shift means liquid cooling demand spreads from "single chips" to the entire whole-rack chain. Overseas Infrastructure Gap Is Evident, Domestic Manufacturers Are Expanding Overseas Smoothly According to TrendForce data, the global AI data center liquid cooling penetration rate rose from 14% in 2024 to 33% in 2025, and is expected to reach 40% in 2026. In terms of the global liquid cooling market's regional structure, North America has the largest volume and Southeast Asia has the fastest growth, and these are precisely the regions where local capacity is tightest and acceptance of Chinese supply chains is relatively high. Overseas leaders' orders are growing rapidly, with the Middle East and Southeast Asia as new main lines. Domestic manufacturers' orders are being validated and implemented, and overseas expansion is progressing smoothly. National Standards Land to Advance Mass Production, Gross Margin Recovery Is Imminent 2026 is a year for the concentrated implementation of the liquid cooling standards system, pushing the industry from "project-based customization" to "standardized mass production" and clarifying issues of incompatible interfaces and differing specifications. The liquid cooling equipment industry continued its downward trend in 2026H1, and it is expected that in the second half, as scale effects emerge, gross margins are expected to recover, with the focus of competition shifting from "whether supply is possible" to "whether system integration is possible." Risk Factors: Risk of demand falling short of expectations; risk of profit pressure due to intensified competition; risk of Rubin mass production and delivery pace falling below expectations; risk of technology route iteration; risk of overseas geopolitical and trade uncertainty.