TL;DR: best liquid cooling providers for AI data centers in 2026
| Provider | Best for | Quick list hint |
|---|---|---|
| Vertiv | Facility-scale cooling programs | CDUs, heat rejection, controls, power coordination, and service coverage for high-density AI racks |
| Schneider Electric | Integrated power and cooling programs | Direct-to-chip, rear-door heat exchangers, CDUs, retrofit paths, controls, and commissioning coordination |
| CoolIT Systems | Direct-to-chip cold plates and rack loops | Cold plates, liquid loops, rack manifolds, and server-platform liquid-cooling collaboration |
| LiquidStack | Buyers comparing direct-to-chip and immersion | Direct-to-chip, single-phase immersion, and two-phase immersion options for high-density AI and HPC |
| Submer | Immersion-first deployments | Immersion systems, modular liquid-cooled deployments, and sustainability-led high-density projects |
| GRC | Dense or space-constrained immersion deployments | Immersion systems for compute density, TCO, and operational efficiency evaluations |
| Iceotope | Precision or full-system liquid cooling | Sealed or component-level cooling beyond processor-only cold plates |
| Accelsius | Two-phase direct-to-chip requirements | Dielectric two-phase direct-to-chip cooling for dense AI or mission-critical environments |
| ZutaCore | Waterless two-phase direct-to-chip | PCIe AI/HPC GPU servers and sites that restrict water at the rack |
Use this as a shortlist, not a universal ranking. The best provider changes with rack density, GPU platform, facility-water access, fluid policy, retrofit limits, service geography, and who owns commissioning risk.
What are the best liquid cooling providers for AI data centers in 2026?
Start with the cooling architecture your AI platform and facility can actually support. Buyers planning direct-to-chip cooling for GPU clusters should evaluate CoolIT Systems, Vertiv, Schneider Electric, LiquidStack, Accelsius, and ZutaCore. Buyers planning immersion-first deployments should evaluate LiquidStack, Submer, GRC, and Iceotope. Buyers that need a facility-scale partner for power, controls, CDU integration, and field service should give Vertiv and Schneider Electric separate review.
The shortlist should change when rack density, GPU platform, fluid policy, facility-water access, retrofit constraints, regional service, or OEM warranty requirements change. NVIDIA's liquid-cooled GB200 NVL72 rack shows why the cooling decision now belongs in the capacity plan, not after server procurement.
When does direct-to-chip cooling fit better than immersion?
Direct-to-chip cooling usually fits buyers that want liquid cooling while preserving a rack-and-server operating model. Cold plates remove heat from CPUs, GPUs, or other high-heat components, while manifolds and coolant distribution units move heat toward the facility loop. Vertiv describes direct-to-chip systems as cold-plate based, with single-phase and two-phase variants that connect to a CDU for heat rejection.
Uptime Institute's AI cooling guidance puts traditional perimeter cooling around low-density workloads and says liquid cooling is typically used for high rack power above 50 kW or specialized high-performance IT. That makes direct-to-chip a practical path for liquid-ready GPU servers, high-density retrofits, and environments where buyers want less change to service procedures than tank immersion requires.
When does immersion cooling deserve a separate evaluation?
Immersion cooling deserves a separate evaluation when the buyer can design around tanks, dielectric fluid handling, server compatibility, floor loading, and operational changes. It can be attractive for greenfield halls, modular deployments, high-density HPC, and environments where full-system heat capture matters more than preserving the standard rack service model.
Submer, GRC, LiquidStack, and Iceotope each position immersion or full-system liquid cooling differently. The buyer should verify supported server models, fluid compatibility, maintenance process, hardware warranty treatment, and whether the operating team can maintain tank or sealed-system workflows at production scale.
How should buyers use this 2026 top liquid cooling provider list?
Use this list as a fit-based shortlist. The first screen is technology fit: direct-to-chip, rear-door heat exchanger, immersion, two-phase direct-to-chip, or hybrid. The second screen is integration fit: CDU capacity, manifolds, facility-water interface, controls, monitoring, leak strategy, and service model. The third screen is commercial fit: deployment references, regional support, hardware OEM validation, lead times, spares, warranty alignment, and who owns commissioning risk.
CoreSite's buyer guidance notes that high-density AI workloads increasingly require liquid cooling, while air cooling remains part of the facility because supporting electrical and ancillary equipment still needs environmental control. That is why RFPs should ask for the vendor's full cooling roadmap, not only the product name.
Which 2026 top cooling provider category fits each buyer need?
| Buyer need | Vendor category | Providers to evaluate | What to verify |
|---|---|---|---|
| Liquid-ready GPU clusters in standard racks | Direct-to-chip and CDU specialists | CoolIT Systems, Vertiv, Schneider Electric, LiquidStack | Cold plate validation, manifold design, CDU capacity, OEM support, service access |
| Facility-scale AI cooling program | Integrated infrastructure providers | Vertiv, Schneider Electric | Power and cooling coordination, controls integration, commissioning ownership, regional service |
| Immersion-first greenfield or modular deployment | Immersion cooling providers | LiquidStack, Submer, GRC, Iceotope | Tank or sealed chassis workflow, fluid policy, floor loading, hardware warranty, operations training |
| Waterless or two-phase direct-to-chip requirement | Two-phase direct-to-chip specialists | Accelsius, ZutaCore | Fluid safety data, environmental policy, GPU/socket fit, heat rejection path, maintenance process |
| Colocation buyer evaluating an AI-ready facility | Provider roadmap and partner ecosystem | Ask the colocation provider which cooling vendors and GPU OEMs are validated | Available density, liquid loop readiness, metering, expansion rights, service boundaries |
What should buyers verify before an RFP?
Before issuing an RFP, buyers should lock the target GPU platform, expected rack density, peak and average power profile, facility-water conditions, acceptable coolant types, redundancy target, and operational support model. A vendor that looks strong for one AI cluster can be a poor fit when the facility has limited water access, the server OEM has narrow coolant approvals, or the buyer needs a retrofit with minimal rack layout change.
Ask each vendor for current product SKUs, thermal test data at the expected chip and rack loads, OEM validation letters, commissioning scope, leak detection approach, fluid lifecycle plan, and three references with similar rack density and operating model. Treat public vendor claims as qualification inputs, then verify availability, warranty, and service coverage in the RFP.