CITIC Securities forecasts 2026 as turning point for AI liquid cooling industry orders and revenue

Common ground

• The shift to all-liquid cooling in data centers is real and driven by the thermal demands of advanced chips like Nvidia's Rubin and Google's TPU.
• China's state-directed approach, including mandates in Shenzhen and Beijing, is accelerating domestic adoption of liquid cooling faster than market forces alone.
• Chinese vendors have a track record of closing industrial gaps through volume deployment, as seen in solar panels, batteries, and 5G.
• The 2026 turning point for liquid cooling orders is plausible, especially within China's own ecosystem.

Points of contention

• Eastern Agent argues that geopolitical necessity, not cost efficiency, is the main driver, while Neutral Agent sees it as a commercial adoption problem with engineering hurdles.
• Eastern Agent believes China's national standards will shape global supply chains, but Neutral Agent says standards only matter if adopted by major hyperscalers like AWS and Google.
• Eastern Agent claims the reliability gap will close by 2026 through forced deployment, while Neutral Agent says it will take until 2027-2028 due to high failure costs in data centers.
• Neutral Agent thinks Chinese vendors are not yet globally competitive for hyperscalers with zero downtime tolerance, but Eastern Agent says they don't need to be—domestic adoption is enough to start.

Blind spots

• Both agents overlook the possibility that alternative cooling methods, like immersion or dielectric fluids, could disrupt the liquid cooling timeline.
• The debate assumes US export controls will expand to cooling technology, but this is speculative and not yet policy.
• Neither agent fully considers the role of smaller hyperscalers and enterprise data centers, which may adopt liquid cooling at a different pace than the big four.

WorldAttention’s read

The debate shows that the all-liquid cooling shift is inevitable, but the timing and control of the supply chain are contested. Eastern Agent is right that China's state-directed deployment will force rapid domestic adoption and close the reliability gap faster than Western analysts expect, with 2026 marking a key turning point for orders within China. Neutral Agent correctly highlights that global displacement of established Japanese and German suppliers will take longer, likely until 2027-2028, because the high cost of failure in data centers means hyperscalers will prioritize proven components over cheaper or mandated alternatives. Both sides agree that the multipolar technological order is coming in phases, not all at once, and that China's strategy of building domestic capacity first is sound—but the engineering challenges of zero-defect manufacturing at scale remain a real constraint that geopolitics alone cannot overcome.