Liquid cooling TAM is growing faster than any adjacent data center infrastructure market because AI rack power density is crossing the air-cooling threshold simultaneously across all hyperscalers. The base case reflects 150kW/rack AI deployments driving cold plate and CDU demand through 2028, followed by immersion cooling reaching commercial scale by 2029. Bear case assumes hyperscalers slow AI capex in 2027–2028 on utilization concerns, reducing new rack deployments and cooling demand. Bull case assumes 400kW+ racks become the standard architecture by 2028, accelerating immersion cooling adoption and expanding TAM beyond current projections.
“NVIDIA GB300 NVL72 racks draw 120–150kW per rack — beyond the thermal threshold where air cooling is physically viable. Direct-to-chip liquid cooling is now a system requirement, not an option. Vertiv guided liquid cooling revenue growing >60% Y/Y in 2026 as hyperscaler deployments scale.”
“Omdia forecasts cold plate shipments surging from 8 million units in 2025 to 356 million by 2030 — a 44× increase in five years. The supply chain for cold plates, manifolds, and coolant distribution units (CDUs) does not exist at that scale today. Lead times for rack-level CDUs now extend 9–12 months.”
“Quick-disconnect couplings (QDCs) — the fittings connecting cold plates to facility cooling loops — are concentrated among four Western suppliers: Stäubli, Eaton, CPC, and Parker-Hannifin. Chinese competitors face stringent certification timelines. A single missing QDC can delay an entire rack installation. Supply qualifications are running 12–18 months behind demand.”
“Immersion cooling adoption accelerating for 400kW+ density deployments. GRC (Green Revolution Cooling) and Submer securing hyperscaler pilots. Full immersion eliminates cold plates and QDCs entirely — simpler per-rack but requires new facility infrastructure. Projected 15–20% of new AI data center capacity built immersion-ready by 2028.”
“Only four PSU (power supply unit) vendors are currently certified by NVIDIA for next-generation rack deployments, underscoring how constrained the certified cooling-adjacent supply chain is. Certification backlogs across CDU, manifold, and PSU suppliers mean even well-capitalized customers cannot accelerate deployments beyond the qualification timeline.”
Air cooling maxes out at roughly 30–40kW per rack. NVIDIA GB300 NVL72 racks draw 120–150kW. Every AI cluster built at this density requires liquid cooling — there is no engineering workaround. This is not a preference or cost optimization; it is a physical necessity. Every hyperscaler deploying AI compute at scale becomes a liquid cooling customer.
AI racks 2026: ~120–150kW per rack (NVIDIA GB300 spec)
Air cooling limit: ~30–40kW/rack → liquid required above this
Global AI rack deployments 2026–2028: ~3–5M racks estimated
3M racks × $5K–8K avg liquid cooling BOM per rack = $15–24B
+ CDU / manifold / facility loop infrastructure × 1.3 multiplier
→ $15–25B cumulative cooling hardware demand 2026–2028
Cold plate shipments are forecast to grow from 8 million units in 2025 to 356 million by 2030. The supply chain — precision-machined copper or aluminum cold plates, brazed fittings, thermal interface materials — simply does not exist at 40× current scale. Manufacturing a cold plate requires CNC machining tolerances tighter than most industrial components. New capacity takes 18–24 months to qualify.
Cold plate units: 8M (2025) → 356M (2030) per Omdia = 44× growth
Average cold plate ASP: ~$40–60 for server-grade unit
356M × $50 avg = $17.8B annual cold plate TAM by 2030
Cumulative 2026–2030 above baseline demand: ~$8–16B incremental
→ Supply chain constraint compounds: new CNC capacity ~18 months to qualify
Single-phase and two-phase immersion cooling submerges entire servers in dielectric fluid, eliminating cold plates and airflow entirely. At 400kW+ rack densities — the roadmap for post-2028 AI clusters — immersion is the only viable approach. GRC, Submer, and LiquidStack are securing hyperscaler pilot contracts. If any major hyperscaler standardizes on immersion for a new campus, it creates a winner-take-most market for facility design and dielectric fluid supply.
Immersion cooling ASP per rack: ~$15K–25K (vs. $5–8K for cold plate)
If 15% of 2029–2030 AI racks (~500K racks) go immersion:
500K × $20K avg = $10B incremental vs cold plate baseline
+ Dielectric fluid recurring revenue: $500–800/rack/year × 500K = $250–400M/yr
→ $5–12B net incremental TAM vs. cold plate baseline