# As AI Scales, Can Traditional Cooling Keep Up? — retained record

> Retained news record from news:datacenterknowledge.

Canonical: https://highsignal.app/data/records/0183bb692b31e23a

[← sources](https://highsignal.app/data)


source: [news](https://highsignal.app/data/news) (news:datacenterknowledge)  published 2026-06-04 14:30 ingested 2026-06-05 09:34

[original source ↗](https://www.datacenterknowledge.com/cooling/as-ai-scales-can-traditional-cooling-keep-up-)

## retained text



```
Insight and analysis on the data center space from industry thought leaders.
As AI Scales, Can Traditional Cooling Keep Up?
As AI workloads push rack densities higher, operators are rethinking cooling strategies as power demands, water use, and infrastructure limits collide.
AI and high-performance computing (HPC) are reshaping what "normal" looks like inside a data center. Accelerator servers built for AI training and inference draw far more power than conventional CPU systems and often operate near peak utilization for extended periods.
That sustained load makes heat removal a critical factor in how quickly operators can add capacity. When temperatures exceed what silicon can tolerate, performance drops through throttling and reliability risks increase. AI workloads also create highly dynamic heat patterns across GPU dies, with hotspots that can exceed average heat flux by several multiples and shift in milliseconds.
At the same time, external scrutiny is growing. Power availability already influences where new capacity can be built and how quickly it can come online. The International Energy Agency (IEA) projects global data center electricity consumption could reach roughly 945 terawatt-hours by 2030, more than double 2024 levels, with AI serving as a major driver.
Water availability is increasingly part of the discussion, particularly in regions facing drought and competing resource demands.
As demand grows, a key question emerges: can today's cooling methods scale alongside it?
Air cooling remains effective in many environments, but its limitations become more apparent as rack power increases.
Removing more heat with air requires moving significantly larger volumes of air, increasing fan energy consumption and tightening operational tolerances. AI workloads often sustain high utilization, leaving less room for variation in airflow or power conditions before throttling occurs. As a result, many operators are turning to liquid cooling for higher-density deployments.
Liquid Cooling's Trade-Offs
Liquid cooling removes heat closer to the silicon, reducing dependence on room airflow and ambient temperature. Many deployments use water-based direct-to-chip systems, which can perform well but introduce additional operational considerations.
Water near IT equipment raises the consequences of leaks and downtime events. Operators must also manage water quality, corrosion control, and system maintenance over time. Some estimates suggest a 100 MW facility can consume roughly 1.1 million gallons of water per day, depending on cooling design and local conditions.
In Europe, reporting requirements tied to the Energy Efficiency Directive require larger operators to disclose energy and water metrics, bringing cooling choices into broader conversations about sustainability and community impact.
The Growing Role of Waterless Cooling
These pressures are prompting interest in waterless cooling approaches. In this context, "waterless" refers to removing water from the data hall and reducing reliance on evaporative cooling, which drives water consumption in many designs. Typical approaches use closed-loop systems, dry heat rejection where practical, and physical separation between facility water systems and IT equipment.
One approach is two-phase direct-to-chip cooling.
A cold plate sits directly on the processor and contains a dielectric heat-transfer fluid. As the processor heats up, the fluid boils at a controlled temperature and carries heat away as vapor. The vapor then condenses and returns to the cold plate, creating a continuous cycle.
Because phase change performs much of the heat-removal work, these systems may require lower flow rates than single-phase designs that depend primarily on circulating larger volumes of liquid.
Advocates argue this approach can reduce pumping requirements, maintain more stable chip temperatures under changing workloads, and avoid some of the water-treatment and corrosion-management challenges associated with water-based systems.
As rack power rises, distribution architecture also becomes more important. Shared distribution systems serving multiple racks can reduce duplicated infrastructure while maintaining rack-level monitoring and control.
This is particularly relevant for retrofit projects. Most existing facilities were designed around air cooling and offer limited flexibility for major infrastructure changes. Introducing liquid cooling first in the highest-density zones can extend the useful life of existing facilities while creating potential opportunities for heat reuse.
Industry adoption continues to expand. In Uptime Institute's 2025 Cooling Systems Survey, 22% of organizations reported using direct liquid cooling, with higher rack densities cited as a primary driver.
As liquid cooling becomes more common, attention is likely to shift toward operational maturity, including standard interfaces, telemetry, and service models designed for AI, HPC, and colocation environments.
Planning for Higher-Density Infrastructure
AI is increasing chip and rack power requirements faster than many facilities can be rebuilt. As a result, cooling decisions are becoming a central part of infrastructure planning alongside power availability, water use, and operational efficiency.
For new facilities, that means evaluating cooling approaches that can support future generations of accelerators without creating unsustainable resource demands. For existing facilities, it may mean introducing liquid cooling gradually in the highest-density environments while adapting operational procedures and monitoring systems to support those deployments.
How operators address cooling will play a significant role in determining how quickly new AI infrastructure can be deployed and how it is viewed by regulators, customers, and local communities.
```



document fetched 2026-06-05 09:34