Operators are reporting on single-phase reliability problems

Last month, leaders from all corners of the data center industry gathered in Dallas for the ninth-annual High-Density AI Data Center Cooling & Thermal Management Summit. What was on the agenda? Observed pain points from past and present liquid cooling deployments.

Session takeaways included how to address “localized hotspots and uneven rack-level heat distribution,” “infrastructure vulnerabilities,” “water constraints,” at-scale leak risks, and facility operators’ “cognitive overload” coupled with “workforce loss.” Beyond these operational hurdles, one existential concern was raised as one of the program’s key themes: “Can cooling architectures keep up with AI workload demands?”

Enough time has passed for liquid cooling—particularly single-phase direct-to-chip, the current incumbent on the market—to go from an up-and-coming solution to broad adoption. That time has allowed facility operators to observe the aftereffects of these high-density deployments. Clearly, issues persist; and these issues are so prevalent that an entire industry summit was dedicated to discussing them.

Enough time has passed for liquid cooling—particularly single-phase direct-to-chip—to go from an up-and-coming solution to mass market adoption. That time has allowed facility operators to observe the aftereffects of high-density deployments. Clearly, issues persist; and these issues are so prevalent that an entire industry summit was dedicated to discussing them.

This is also why two-phase D2C cooling has gained so much attention. Yes, it has a proven runway to ensure mission critical cooling keeps up with AI’s breakneck thermal demands. But for those preparing to transition from 1P D2C to what’s needed next, 2P offers one essential advantage: it can eliminate the specific pain points that 1P operators keep running into. And as we’ll soon see, those pain points can cause huge financial penalties in the near term.

Lack of reliability is affecting ROI

Recently, Uptime Institute published its 2026 Annual Outage Analysis, ranking the key culprits for costly operational downtime. For the past several years, cooling has been consistently ranked as a leading cause for outages deemed “major”; this year, the report points to worsening pressure from high-density AI workloads as a reason why cooling-related downtime is increasing. Alongside this growing risk are the growing chances of financial disaster: over half of major outages now cost more than $100,000, and about one in five exceed $1 million.

1P D2C currently holds most of the liquid cooling market, which means it’s carrying most of this financial exposure. But why does this costly risk of downtime still exist? It’s largely the same reason why 1P is burdened with so many of the maintenance concerns we mentioned above; namely, the water flowing through its tubes.

The growing checklist surrounding water chemistry

We’ve written elsewhere about the hidden costs that 1P maintenance can inflict on unsuspecting facility operators. Now, you don’t have to take our word for it—water treatment companies have started to document their own difficulties maintaining 1P cooling systems.

In an article for Water Technology Online, author Pete Elliott makes one core observation: installation completion doesn’t always equal operational readiness. In fact, a 1P system can pass any installation checklist and still carry water-chemistry risks into production.

Water-based direct-to-chip loops are unusually exposed to risks due to their own design. Flow passages inside cold plates are often narrower than a human hair, and CDUs commonly rely on filters in the 25-75 micron range. Those constricted spaces easily allow debris left behind by imperfect flushing (an essential aspect of 1P maintenance) to settle and foul tubing. Left unresolved, that stray debris can promote microbial growth and further corrode the system—even before it runs at full capacity.

In the article, Elliott offers a checklist for “genuine readiness.” It includes the following:

• Have water systems been cleaned, flushed, passivated, and supervised by personnel with commensurate experience?
• Has water chemistry been stabilized?
• Have suspended solids and construction contaminants been reduced to meet pre-established quality requirements?
• Have monitoring and sampling points been set up and validated?

Consider just how much of an ongoing process even part of this checklist implies. Consider, then, how much of time and effort and OpEx will be allocated towards compliance and maintenance. All this must be considered when the industry’s talent gap is widening, and an individual operator’s “cognitive workload” is increasingly strained.

All told, this article and the talks surrounding the Dallas summit point to one clear lesson: 1P systems require sustained, expert-level water treatment discipline indefinitely. Any lapse in that discipline becomes a reliability problem on the scale operators are now reporting. If that reliability fails, water-based leaks occur—and millions of dollars’ worth of damage to critical AI infrastructure can occur in mere moments.

1P systems require sustained, expert-level water treatment discipline indefinitely. Any lapse in that discipline becomes a reliability problem on the scale operators are now reporting. If that reliability fails, water-based leaks occur—and millions of dollars’ worth of damage to critical AI infrastructure can occur in mere moments.

2P paves a risk-free thermal runway

By now, we’ve hopefully presented enough evidence of operators noticing the unfortunate side effects of 1P deployments. Luckily, they’re also noticing that 2P directly eliminates most of 1P’s risks and hassles, offering a compelling path to greater thermal potential without the headache of complex operational overhead.

In a Storage Review podcast, Tim Shedd, former Senior Distinguished Engineer at Dell, had this to say about 2P’s key advantages over the current liquid cooling landscape:

If we think of the issues of water—water quality, leaking, all this stuff—they’re dramatically minimized [with 2P]. I think there’s a lot of potential in the enterprise space…if they don’t want to take the risk of water in their rack, 2P might be a great answer to them. The same thing could be true for neoclouds: it’s a risk mitigator.

It’s 2P’s lack of water that avoids so many of the problems 1P creates post-deployment. Let’s revisit that litany of issues explored during the Dallas summit, and see exactly how 2P addresses each one:

• “Localized hotspots and uneven rack-level distribution.” 2P relies on isothermal—or uniform—heat distribution, meaning it avoids the uneven heat spread (and risk of components being bent or warped) that naturally occurs with 1P.
• “Infrastructure vulnerabilities.” Beyond 1P’s risk of corrosion due to water chemistry, its reliance on exaggerated flow rates to cool AI workloads can cause erosion. 2P’s dielectric refrigerant causes zero microbial growths, and its ability to boil means it requires one-third of 1P flow rates to cool the same workloads.
• “Water constraints.” 2P is an entirely waterless solution. Plus, its 36-56% greater energy savings vs. 1P also slashes local water consumption—a key reason for current community pushback.
• At-scale leak risks. 2P’s dielectric refrigerant is non-conductive. If leaks occur, they cause zero damage to electrical equipment.
• Operators’ “cognitive overload” and the industry’s “workforce loss.” 1P is proven to require a whole new workforce to maintain its peak performance. 2P is instead a set-it-and-forget-it solution that outruns AI’s thermal demands with worry-free operations.

This year, the consequences of 1P deployments have become a central topic of industry conversations. So, ask yourself: do you want to continue worrying with colleagues about what water-based cooling could do to your data center? Or would you rather transition to 2P: a proven solution that eliminates near-term risks and extends a long-term thermal runway?