CITIC SEC: Liquid cooling industry faces order and performance realization period, seize three main opportunities.

CITIC SEC released a research report stating that the liquid cooling industry is transitioning from thematic expectations to the phase of order fulfillment and performance realization. The focus is on supply chain companies with production capacity flexibility, customer certification, and core component capabilities. In the short term, the OEM chain has a high performance elasticity, with the mass production of GB200/GB300 leading to tight capacity for Taiwanese and international leading companies, while mainland companies rely on precision processing and assembly to take on overflow orders. The supply chain for independent brands has the strongest long-term growth potential, positioning itself in the CSP, ODM, and domestic GPU ecosystem with certification assets and customer interface, currently in the prototype and small batch validation phase. Horizontal new product categories provide differentiation opportunities, with low value but broad and easy entry for products like power distribution systems, liquid-cooled optical modules, etc. It is suggested to seize the opportunity along the three main lines of "OEM chain performance elasticity, independent brand certification breakthrough, and horizontal new product category volume release." Key points from CITIC SEC are as follows: AI computing power infrastructure is transitioning from server-level cooling to rack-level liquid cooling era. With platforms like GB200, GB300, Rubin continuously increasing single cabinet power, chip density, network bandwidth, and memory capacity, the cooling bottleneck is no longer just temperature control for single GPU/CPU, but the whole cabinet's heat load management capability. Liquid cooling has thus upgraded from single-point cold plate solution to basic infrastructure capability covering chips, power supply, network, optical modules, and cabinet systems. NVIDIA's Rubin platform has achieved 100% liquid cooling for the entire system, driving cloud vendors and data center operators to accelerate the switch to liquid cooling architecture; meanwhile, the increase in power density of ASICs developed in-house by Google, Meta, Amazon, etc., is also pushing for more efficient cooling solutions for customized AI servers. Liquid cooling demand is expanding from core computing chips to various heat sources throughout the whole cabinet. By 2024, the penetration rate of AI liquid cooling has reached 14% (according to TrendForce), signaling the accelerated adoption of liquid cooling in AIDC, with the market space for global AIDC servers' liquid cooling projected to reach $21.8 billion by 2027 (mainly for calculating the chip end). Apart from GPU/CPU, components like Busbar power distribution systems, optical modules are becoming new areas for liquid cooling. In terms of power distribution, liquid-cooled Busbars can significantly improve the cabinet's power-carrying capacity, with liquid cooling rack power reaching up to 750kW under 48V voltage (compared to around 140kW for air cooling). It is estimated that by 2026, the global AI data center liquid-cooled DC busbar component market sales volume will be around $230 million, with a projected growth to $640 million in 2032, with a CAGR of around 27% (based on QYResearch forecasts). At the optical module end, with 800G OSFP power consumption reaching 25W-30W and 1.6T DR8 DSP peak power reaching 40W-50W, the penetration rate of liquid cooling is expected to increase from about 5% at the 800G stage to around 40% at the 1.6T stage. By 2030, the market size of liquid-cooled optical modules is expected to exceed $6.3 billion (prediction from Zero-Krypton 1+1 WeChat public account). The core barriers of the liquid cooling industry chain come from precision manufacturing, reliability verification, and scale delivery capabilities. Products like cold plates, CDUs (Cooling Distribution Units), Manifolds, UQDs (Quick Disconnects), pipelines, and liquid-cooled cabinets involve complex processes like microchannel processing, vacuum brazing, seamless welding, airtight testing, thermal cycling aging, etc., requiring high equipment investment, process teams, quality control, and customer certification. The industry is not simply about assembling for scale-up; stable yield rates and long-term delivery experience determine whether a company can enter the mainstream supply chain. Taiwanese companies maintain a leading position in the liquid cooling chain of NVIDIA and North American CSPs due to early certification, deep customer relationships, and overseas production layouts. The global liquid cooling supply chain is currently dominated by Taiwanese manufacturers, with mainland manufacturing accelerating to take on capacity. North America remains the largest region for liquid cooling demand, with a market share of about 65%, followed by Southeast Asia, mainland China, Europe, and Japan/Korea at approximately 15%, 10%, 5%, and 5%, respectively (estimation from Zero-Krypton 1+1 WeChat public account); on the supply side, Taiwanese manufacturers still have a strong advantage in key components like cold plates, CDUs, with an overall market share of about 50%-60% (estimation from Zero-Krypton 1+1 public account). However, as Taiwanese capacity reaches saturation, the proportion of mature liquid cooling components outsourced overseas has exceeded 50%. Mainland companies, with manufacturing clusters in East China and South China, are taking on overflow capacity in areas like CNC machining, welding, piping, joints, assembly, and testing, gradually increasing their participation in domestic intelligent computing centers and domestic GPU ecosystems. The liquid cooling industry is entering the phase of order fulfillment, with mainland companies breaking through the Taiwanese dominance through independent brands, OEM support, and new product categories. Although Taiwanese liquid cooling companies maintain a monopoly position and revenue growth far ahead of mainland Chinese manufacturers due to their first-mover advantage, the industry is now entering a phase where orders and performance realization are becoming crucial with the full-scale production of Rubin and Google TPU V8. Mainland Chinese manufacturers are adopting three parallel approaches to break through the Taiwanese dominance: positioning their independent brands in CSPs and domestic GPU ecosystems for long-term growth; taking on overflowing capacity from Taiwanese manufacturers in the OEM chain to release short-term performance elasticity; and entering new product categories like optical modules, Busbars for differentiation to expand customer coverage, presenting an overall pattern of OEM performance realization, brand solidification, and differentiation breakthrough in new product categories. Risk factors: Lower-than-expected demand for AI computing power; risks associated with technological iteration and substitution; industry competition intensifying leading to a decline in profitability; risks associated with core customer certification and entry barriers; geopolitical and overseas policy risks.