Cooler Master Announces 2.5MW In-Row CDU for AI Data Centers

Cooler Master has announced a 2.5MW liquid-to-liquid in-row Coolant Distribution Unit (CDU, a device that circulates coolant between facility water and server cold plates) for liquid-cooled data centers. Mass production is targeted for Q4 2026. The company disclosed no customers, no pricing, and no deployment commitments.

The Taiwan-based firm, founded in 1992 and historically a supplier to the PC gaming market, is pushing further into AI infrastructure cooling. Over the summer it announced an expansion into the data center segment built around Nvidia's MGX modular reference design architecture.

What are the stated specifications?

The unit delivers up to 2.5MW of cooling capacity at 990 GPM (gallons per minute of coolant flow) with a 4-5°C approach temperature difference — the gap between the leaving coolant temperature and the entering facility water temperature. Each unit consumes 40kW to operate. The company said the system lets operators "support larger AI clusters, higher rack densities, and future infrastructure growth without compromising reliability."

That reliability claim arrived without published redundancy data, availability figures, or field operating history. A vendor entering the rack-scale segment from consumer computing has no track record at this capacity class.

How far does the system scale?

The CDU uses a dual-system architecture with load balancing and centralized management. It supports up to 32 interconnected CDUs, scaling to 80MW of total cooling capacity.

The company offered no details on what happens to the 80MW figure beyond 32 units, nor on failover behavior when one unit in an interconnected group drops offline.

The new CDU sits within a broader portfolio for AI infrastructure:

• Cold plates, manifolds, and filling and commissioning tools
• Liquid-to-liquid CDUs starting from 150kW, in both in-rack and in-row configurations
• A 300kW Air-Assisted Liquid Cooling (AALC) CDU, already in production
• A 30kW in-rack liquid-to-air model, targeted for production in Q4 2026

The liquid-to-air systems address operators hosting liquid-cooled servers inside existing air-cooled facilities — a retrofit path that avoids facility water loop upgrades.

What is Cooler Master not saying?

The announcement is a product roadmap, not a shipment. Both headline items — the 2.5MW CDU and the 30kW in-rack unit — carry the same Q4 2026 target, roughly two years out. The 4-5°C approach temperature claim lacks stated test conditions, and the 40kW operating draw is not broken down by component. Two-phase cooling work is mentioned without a capacity figure, timeline, or product name.

For comparison, the new flagship unit delivers more than eight times the capacity of the 300kW AALC unit already in production — a measure of how far the company is reaching beyond its current shipping portfolio.

No pricing was disclosed. No reference customers or trial deployments were named.

Why does this matter to the market?

Rack densities for AI clusters now exceed what air cooling alone can handle, and CDU capacity has become a procurement bottleneck. Cooler Master's entry adds another supplier to a segment where established thermal vendors already compete at multi-megawatt scale. Whether a consumer-cooling heritage translates into data center-grade reliability will be tested in Q4 2026, and not before.