AI data‑centre cooling becomes a headache; Google reportedly in talks to buy Chinese liquid‑cooling systems

Google reportedly seeking Chinese cooling tech as AI racks get hotter

It has been reported that Google is in talks with Chinese suppliers to buy liquid‑cooling systems for its AI data centres. The talks, reportedly driven by the skyrocketing power density of modern AI accelerators, underscore a problem that many cloud operators now face: chips are getting hotter faster than air cooling can keep up. Simple solution? Move to liquid — but not without trade‑offs.

Why liquid cooling matters — and why China

High‑performance AI chips can consume hundreds of watts per chip, and when packed into dense racks the heat and energy efficiency problems multiply. Liquid cooling (direct‑to‑chip or immersion) can slash power usage for cooling and enable higher compute density, boosting throughput per square metre. Chinese suppliers have invested heavily in these systems and scaled manufacturing quickly, making them attractive on price and delivery timelines to hyperscalers racing to expand capacity.

Geopolitics and procurement risk

But buying from Chinese hardware vendors is not just an engineering choice. It has been reported that U.S. regulators and security reviewers are watching closely as Washington tightens export controls and scrutinises foreign technology linked to critical infrastructure. Is this simply pragmatic procurement — or a geopolitical vulnerability? Reportedly, such transactions could trigger reviews under U.S. national‑security frameworks and raise concerns among allies about supply‑chain trustworthiness.

Market implications

If Google — or other Western cloud giants — moves forward, Chinese cooling vendors could gain a stronger foothold in global AI infrastructure markets. That would reshape vendor dynamics while intensifying regulatory debates over where and how critical data‑centre components are sourced. For now, details remain limited and Google has reportedly declined to comment. The bigger question lingers: can engineering urgency overcome geopolitical caution?