If you walk mission-critical facilities in 2026, one pattern stands out immediately.
The operators quietly delivering the best PUE, WUE, and OpEx numbers aren’t chasing colder process chilled water (PCW). They’re strategically raising it.
And in multiple hyperscale AI campuses I’ve reviewed this year, this single operational adjustment is delivering seven-figure annual savings — often with zero CapEx and zero impact on uptime or chip performance.
A few years ago, this would have been considered risky. Today, it’s becoming table stakes for anyone running 60–100+ kW racks with direct-to-chip liquid cooling.
Why the Old Rules No Longer Apply
AI has fundamentally changed the thermal game:
- Rack densities routinely exceeding 100 kW
- Widespread adoption of direct-to-chip and CDU-based liquid cooling
- Severe power and water constraints in key markets
- Intense pressure to deploy capacity faster than new infrastructure can be built
Yet many plants still run legacy PCW supply temperatures in the 6–10°C (43–50°F) range — a holdover from comfort-cooling mindsets.
That conservative approach now creates hidden drag: higher chiller lift, reduced free-cooling hours, low ΔT syndrome, excess pumping energy, and higher water consumption.
The New Sweet Spot: 14–20°C (57–68°F) PCW Supply
Leading AI facilities are safely operating PCW supply temperatures in the 14–20°C range. Many are pushing even higher on the facility water side feeding CDUs.
This isn’t theoretical. The physics is straightforward: every 1°C increase in chilled water supply temperature typically reduces chiller energy consumption by 2–3%. Scale that across a 50–100 MW AI load and the numbers become transformative — 20–40% cooling energy reduction is realistic when combined with wider ΔT, optimized flow, and more economizer hours.
Real results I’ve seen:
- Significantly lower compressor work and peak demand
- Thousands of additional free-cooling hours per year
- Reduced cooling tower evaporation (better WUE)
- More stable plant operation and easier heat recovery
[Insert Graphic: Legacy Low-Temp PCW vs. Elevated PCW Strategy – side-by-side comparison with energy & water metrics]
One Facility That Avoided a Multi-Million Dollar Chiller Expansion
A recent AI compute campus was planning a major chiller plant addition. The projected loads looked overwhelming.
After detailed analysis, the real issue wasn’t insufficient tonnage — it was excessively low PCW setpoints inherited from older designs. The GPUs and CDUs didn’t need that cold water.
We executed a controlled, phased increase of ~5°C with continuous monitoring of rack inlet temps, heat exchanger approach, valve authority, and ΔT. Results after full implementation:
- Chiller power dropped dramatically
- Economizer runtime increased substantially
- Chronic low-ΔT issues resolved
- Pump and control stability improved
- Projected annual savings exceeded $1.2M
- No new chillers required. No downtime. Full redundancy maintained.
This story is repeating across the industry right now.
The Consultant Playbook: How to Do This Safely and Profitably
Raising setpoints without proper engineering is where things go wrong. Here are the practical, battle-tested steps the best operators use:
- Aggressively Trend Heat Exchanger Approach Temperatures Supply temperature is only part of the story. Monitor approach ΔT on CDU heat exchangers and facility loops closely. A degrading approach often signals flow imbalance or fouling before rack alarms appear.
- Re-Validate Valve Authority and Actuator Tuning First Valves sized for low-temp, high-flow conditions frequently lose authority or become unstable at higher temperatures. Review sizing, authority ratios (>0.5 ideal), and PID settings before making changes.
- Optimize for Higher ΔT, Not Maximum Flow Stop compensating for inefficiency with brute-force pumping. Target wider ΔT across the system. This compounds savings through chillers, pumps, towers, and piping losses.
- Commission Seasonally, Not Just on Design Day Test transitions: partial load, economizer changeover, high ambient, and humidity swings. Many “stable” systems reveal control hunting or thermal excursions only under real seasonal conditions.
- Treat This as a Reliability Initiative, Not Just Efficiency Done right, elevated PCW reduces compressor cycling, mechanical stress, and control oscillation — often improving long-term uptime margins.
[Insert Diagram: PCW/CDU Loop with Key Monitoring Points + Before/After Energy Chart]
Why This Matters Even More in 2026
Power availability, water restrictions, and grid constraints are now the primary bottlenecks — not raw compute demand. Raising PCW temperatures simultaneously attacks energy cost, water usage, deployment speed, and heat reuse potential.
Higher return temperatures also make district heating or absorption cooling far more viable. Several European and forward-looking U.S. campuses are already monetizing waste heat.
Final Thought
The coldest water is no longer the smartest water.
In the AI era, the smartest operators extract maximum performance and reliability from the systems they already have — through better engineering, not bigger infrastructure.
What PCW supply temperature is your facility running today? Drop it in the comments, along with your biggest AI cooling challenge right now. I read every reply and will share practical feedback where I can.
Martin P. King Mission-Critical Cooling Consultant | Helping operators turn hidden inefficiencies into reliable savings
