Google in talks with China’s Envicool, others to buy data centre cooling systems, sources say

Google's procurement team visited China to discuss buying liquid-cooling equipment from Envicool and other suppliers, signaling procurement activity for AI data-centre infrastructure. JPMorgan projects the AI server liquid-cooling market to rise to >$17 billion in 2026 from $8.9 billion last year, while Envicool reported revenue growth of ~40% in the first nine months and a market valuation of 98 billion yuan (~$14bn). Envicool showcased Google-spec coolant distribution units and has potential pipeline orders for fifth-generation CDUs plus planned capacity expansion in Guangdong, Thailand and the U.S., highlighting tightening supply and growing Chinese supplier importance.

Analysis

The immediate, under-appreciated consequence of tighter component flows for AI infrastructure is margin reallocation across the ecosystem rather than a one-off supply shock. Hyperscalers that can shorten deployment lead times by 6–12 weeks capture incremental revenue sooner and convert that into higher effective gross margins on AI services; conversely, upstream vendors with differentiated engineering lose pricing power as use of modular, standardized coolant and CDU assemblies spreads. Expect a two-speed market: high-volume, low-margin commoditized suppliers gaining share by volume while bespoke-engineering vendors face margin compression and extended receivable cycles.

Regulatory and capacity dynamics are the dominant tail risks. Trade-policy shocks that restrict cross-border procurement could create abrupt rerouting costs (3–6 months of deployment delays) and force cloud operators to subsidize on-shore capacity buildouts, swapping OpEx for multi-year CapEx — a catalyst that benefits banks and equipment financiers but slows near-term revenue realization for service layers. On the other hand, factory adds in low-cost jurisdictions create a 12–24 month oversupply risk that would compress component ASPs by an estimated 15–25% and accelerate consolidation among smaller suppliers.

From a positioning standpoint, this is a classic infrastructure-cost dispersion trade: back cash-rich platforms that internalize or flex procurement (mid-term winners) and use capped-leverage option structures to express asymmetric upside to compute demand. The consensus underweights the duration mismatch — sellers of parts monetize faster than buyers recognize savings — creating a window to buy platform exposure and hedge execution/timing risk via defined-loss option structures rather than naked longs on discrete suppliers.

Key Decisions for Investors

• Long GOOGL equity (6–12 month horizon): allocate 2–3% of NAV. Rationale: platform-level capture of lower marginal Opex and faster go-to-market offsets some capex licking; target +20% upside, hard stop -10%. Size to conviction and hedge with short-dated puts if deployment signals slip.
• Buy NVDA 12–18 month call spread (long ~35% OTM / short ~70% OTM) to express continued AI compute demand with capped risk: max loss = premium, target asymmetry ~3:1 if NVDA sustains adoption curve. Use as a leveraged play on chip-driven rack density growth while limiting capital at risk vs naked calls.
• Small overweight (1–2% NAV) in GS or JPM (12–24 months): play financing/IB upside from AI-capex and cross-border M&A; these names benefit if clients push CapEx onshore and refinance. Risk: macro slowdown that curtails fee pools — trim into rally above +15%.
• Hedge execution/timing risk with a collar on GOOGL (sell 3–6 month calls to finance 3–6 month puts): preserves upside capture of margin tailwinds while protecting against short-term procurement/regulatory delays. Use this approach instead of large naked positions in discrete suppliers.