Critical Flaws in Data Center Cooling Systems Threaten AI Infrastructure

New research warns of technical deficiencies that could jeopardize billions in AI investments

Published on Feb. 15, 2026

Got story updates? Submit your updates here. ›

A comprehensive peer-reviewed technical paper reveals alarming deficiencies in current liquid cooling system design and implementation that could jeopardize the reliability of critical AI infrastructure, as well as the lifecycle of the chips and hardware that fuel AI's performance. The research exposes fundamental flaws in published industry guidance for direct-to-chip (DTC) cooling systems, the technology increasingly relied upon to manage heat from high-performance GPUs powering AI applications.

Why it matters

As artificial intelligence workloads push data center rack densities to unprecedented levels, the stakes are higher than ever. A 2025 benchmarking survey showed that 48% of respondents were currently using direct-to-chip cooling systems, carrying immediate implications for billions of dollars in AI infrastructure investments. The research reveals that current industry standards contain inaccurate information about crucial system components, putting these critical investments at risk.

The details

Among the most concerning findings is a widespread misunderstanding about glycol solutions in cooling systems, leading operators to make choices that compromise both data center performance and energy sustainability. The research reveals multiple problems with glycol use in DTC cooling, including massive performance penalties, chemical degradation risk, false security on bacteria, dye interference, and sustainability concerns. The paper also identifies additional high-risk areas where current practices fall short, such as microchannel fouling risk, incompatible testing standards, material compatibility issues, and fluid dynamics confusion.

  • The research paper, 'TP26-06 Optimizing Lost Heat Transfer in Single-Phase Cold-Plate Liquid Cooling Systems,' was published in February 2026 by The Cooling Technology Institute (www.cti.org).
  • The 2025 annual benchmarking survey by Data Center Frontier (N=172) showed that 48% of respondents were currently using direct-to-chip cooling systems.

The players

Loraine Huchler, P.E., CMC®

Cooling systems expert and author of the research paper from MarTech Systems, Inc.

The Cooling Technology Institute

The organization that published the peer-reviewed technical paper.

Got photos? Submit your photos here. ›

What they’re saying

“Some operators assert that inhibited glycol solutions have higher stability and greater reliability compared to water, but the data tells a different story.”

— Loraine Huchler, Cooling systems expert (The Cooling Technology Institute)

“Unfortunately, inaccurate technical information and unverified claims continue to create risks for data center owners and operators.”

— Loraine Huchler, Cooling systems expert (The Cooling Technology Institute)

“The performance of chip cooling systems has not matched the increased power consumption and heat rejection needed to ensure reliable performance. We're seeing increased adoption of liquid cooling, but without rigorous technical standards, the industry is building on a shaky foundation.”

— Loraine Huchler, Cooling systems expert (The Cooling Technology Institute)

What’s next

The paper calls for fundamental reforms by the AI and data center industry, stating that suppliers, water treatment professionals, and consultants must serve the data center community by creating validated technical information to optimize heat transfer efficiency and system reliability.

The takeaway

This research highlights the critical importance of addressing technical deficiencies in data center cooling systems to ensure the reliability and sustainability of the infrastructure powering the AI industry's growth. Addressing these issues is essential to protecting billions of dollars in AI investments and safeguarding the future of this transformative technology.