Datacenter Cooling Product Landscape — May 2026

Why This Map Looks Different Than It Did 18 Months Ago

GB200/GB300-class racks crossed ~100–130kW power density across 2025–2026, breaking air cooling's thermal envelope and moving direct-to-chip liquid + CDUs from optional to *default* inside NVIDIA reference designs. The competitive question is no longer "liquid or air" — it is who is qualified into OEM server lines today versus who is still hiring core CDU engineers (e.g. nVent reqs 2026-03/04, d834c2ed/13a32eaf) and who is shipping research, not product (Modine two-phase, 51ff86d1).

Product Categories

• **Direct-to-chip single-phase DLC** — cold plates + Coolant Distribution Units mounted at the GB200/HGX server; qualified through OEM channels (CoolIT d50280b0; nVent 21812cac; Asetek b6063821).
• **Full-stack integrated thermal portfolio** — one vendor selling CDU → chiller → CRAH → air-handling as a single loop rather than a point product (Modine 0216d5d8; nVent partial 21812cac).
• **Chassis-level / sealed-loop precision liquid** — sealed enclosure cooling that sits between air and full immersion, enabling hot-swap and edge/regulated sites immersion tanks cannot serve (Iceotope 812c5544; Asetek sealed-loop heritage b6063821).
• **Air-side / evaporative heritage migrating to liquid** — evaporative/air-side incumbents converting an installed base toward CDU/liquid (Munters fb5d9edb).
• **Two-phase & immersion frontier** — >100kW-rack-density approaches that are R&D or roadmap, not field-proven shipped product (Modine two-phase 51ff86d1).

Comparison Table

| Product | Company | Quality Anchor (0–100) | Primary Cooling Modality | Use Case | |---------|---------|------------------------|--------------------------|----------| | Coolant Distribution Units + direct-to-chip cold plates | CoolIT Systems | 75 | Direct-to-chip single-phase DLC | Qualified into Dell/HPE/Lenovo/Supermicro GB200/HGX server lines (d50280b0) | | CDU + Airedale chiller + CRAH + AHU portfolio | Modine Manufacturing | 70 | Full-stack integrated portfolio | Single-vendor whole-loop buyer; second-source vs Vertiv (0216d5d8, bfb7fb9e) | | DCT liquid/CDU line | Munters Group | 65 | Air-side heritage → liquid/CDU | Retrofitting existing air-cooled halls to higher density (fb5d9edb) | | KU:L precision liquid (sealed chassis) | Iceotope | 65 | Chassis-level sealed liquid | Edge / regulated / hot-swap sites immersion can't serve (812c5544) | | CDU + manifold + enclosure bundle | nVent Electric | 50 | Direct-to-chip + immersion + enclosure | Bundled thermal+enclosure where rack integration dominates (21812cac, d834c2ed) | | In-Rack CDU | Asetek | 47 | Sealed-loop in-rack CDU | Sealed-loop reuse of AIO IP; datacenter re-entry (b6063821, 887ea462) |

Differentiation Map

• **Best for OEM-qualified GB200/HGX server cooling (ship now):** CoolIT — already design-qualified at all four major server OEMs, functioning as third-party validation (d50280b0).
• **Best for a single vendor across the entire thermal loop / de-risking Vertiv dependence:** Modine — only fielder of CDU+chiller+CRAH+AHU as one system with a hyperscale sales motion (0216d5d8, bfb7fb9e).
• **Best for edge / regulated / hot-swappable deployments:** Iceotope — sealed chassis serves sites where immersion tanks are physically/operationally impossible (812c5544).
• **Best for converting an existing air-cooled installed base:** Munters — three consecutive earnings signals show customers actively migrating from its air line to liquid/CDU (fb5d9edb).
• **Avoid if you need a field-proven >100kW two-phase product today:** Modine — two-phase is a research-engineer-staffed R&D track, no shipped customer-validated product (51ff86d1).
• **Avoid if datacenter roadmap continuity is a hard requirement:** Asetek — exited datacenter liquid cooling ~2019 and wound down SimSports 2024; above-average reversal risk (68eb2b4b).
• **Avoid if you require named-hyperscaler proof:** nVent — acquisition-assembled (EriksenLab + Trachte), still hiring core CDU/design engineers in 2026, zero customer-win signals (13a32eaf).

Tool: Interactive Comparator

[Link to /tools/datacenter-cooling-compare] — filter by cooling modality, quality anchor (0–100), and OEM/partner design-in status; surfaces the maturity caveat (shipped product vs roadmap vs still-hiring) per row. Spec set here; frontend wired in Phase G.

Reverse-Hype Warnings

**Overpromised.** Modine's "full-stack thermal" leadership rests partly on AI-prior portfolio breadth, and its frontier two-phase story is explicitly roadmap, not a shipped, customer-validated product (51ff86d1) — treat any >100kW two-phase pitch as aspirational. nVent's integrated stack is calibrated at exactly parity (50) with *no cited customer-facing differentiation* versus Vertiv/Boyd/CoolIT (d834c2ed), and its line is acquisition-assembled with ongoing core-engineering reqs (13a32eaf) — "broad bundle" is breadth, not proven performance. Munters' product strength is inferred from aggregate sell-through only, with zero named-customer or case-study evidence (6953d4f6); demand trend is not the same as demonstrated differentiation, and its liquid/CDU line is a transition off air-side heritage, not a lead versus entrenched direct-to-chip specialists (a739c332).

**Underrated.** CoolIT's multi-OEM design-ins are the strongest independent validation in this set (d50280b0); the real risk is not product quality but margin/share compression as Vertiv and hyperscalers vertically integrate CDU manufacturing (305e64ba) — an ecosystem threat, frequently mis-read as a product weakness. Iceotope is discounted mainly because precision liquid is absent from GB200/GB300 reference designs (badefbd0), yet the sealed-chassis edge/regulated/hot-swap niche it occupies is genuinely defensible and not contested by immersion or pure direct-to-chip (812c5544). The honest read across the set: design-in evidence (CoolIT) and structural integration risk (CoolIT, nVent) are more decision-relevant than headline anchor scores.