Cold Plates = Datacenter Razor Blades

Research evidence excerpt

Cold Plates = Datacenter Razor Blades

June 11, 2026

Bottom Line: Cold plates are meaningful recurring revenue for the liquid-cooling business

Central to pushback on the CoolIT acquisition has been whether datacenter hardware dilutes ECL’s razor-razorblade model. Cold

plate refresh cycles are an under-appreciated factor, which may prove the opposite. Our bottoms-up work points to a 15-25 mm

US cold plate install-base by 2030, with chip-generation specificity making reuse across GPU cycles increasingly unlikely.

A roughly 3-year refresh cadence on 10-20% mkt share would translate into $200-700 mm in annualized replacement sales for

ECL before coolant, filters, manifolds, or service.

Together the initial install opportunity + replacement cycle build reconcile to ECL’s ~30% liquid cooling CAGR with room

to spare. By 2030, CoolIT is set to scale toward ~7.5% of ECL sales. More importantly, datacenter liquid-cooling tracks to contribute

~100-200 bps toward ECL’s 5-7% organic growth annual target – above the market’s current credit and enough to matter for the

EPS algorithm.

Figure 1: Multiple rack architectures point to massive cold plate Scale of cold plate install base is massive

install base (cumulative through 2030)

50 GW of US datacenter adds point to 15-25 mm cold

US Datacenter need by 2030 - per EPRI/Epoch 50 GW plates by 2030

in total power requirement EPRI/Epoch’s framing of ~50 GW of incremental US

Power usage effectiveness (PUE) 1.25 datacenter capacity by 2030 equates to ~40 GW of

modern ~1.1-1.2; air-cooled 1.3+; legacy 1.4-1.8 critical IT load delivered to chips, networking, memory,

Critical IT load 40.0 GW and storage assuming modern ~1.25 power usage

delivered to chips, memory, storage, networking effectiveness (PUE). We triangulate that into cold plate