The Hidden Costs of Single-Phase

Last month, data center leaders reeled from NVIDIA CEO Jensen Huang’s announcement that its Vera Rubin supercomputers could be cooled by 45℃ facility water. Beyond boldly asserting that chip-level liquid cooling is now an industry imperative—and confirming that it’s two-phase direct-to-chip’s time to shine—his statement forecasts major shake-ups to the entire data center stack. 

Put simply, this increasing emphasis on chip-level cooling relieves strain on current chillers, opening the door for greater innovation in heat rejection infrastructure. “Innovation” doesn’t just mean adopting a wider diversity of heat rejection (such as dry coolers and cooling towers); it can also mean reducing existing chiller infrastructure—if, of course, your liquid cooling solution is strong enough to allow it.

Does two-phase direct-to-chip (2P D2C) allow this degree of innovation? You bet it does. In fact, it outright encourages it. Recent analysis from Jacobs Engineering reveals that 2P D2C’s superior thermal performance required only 12 fan high temperature chillers in a 10MW data center (vs. 16 fan chillers required for 1P D2C). Because heat rejection counts for the extreme majority of energy consumption in a liquid-cooled facility, any reduction in chiller infrastructure largely contributes to massive energy savings over time.

Of course, those energy savings translate to financial savings for your facility. Our streamlined chip-to-chiller infrastructure is a key reason why we deliver 35% savings in annual OpEx vs. 1P D2C, according to Jacobs. But it isn’t the only reason we offer better savings vs. single-phase.

Here’s the truth: while it’s easy to grasp the conceptual simplicity of water, single-phase carries a host of hidden terms and conditions that many may overlook prior to adoption. All of these “hidden Ts and Cs” carry a profound risk of operational burden and a bloated OpEx. Those eager to get their feet wet in the world of liquid cooling could ignore these single-phase warning signs at their own peril—and get nickel-and-dimed to the tune of millions of dollars for years to come.

Here’s just a few of the hidden costs that 1P D2C carries:

Higher Pumping Power Requirements

Unlike 2P D2C, which absorbs more heat by boiling its liquid, 1P D2C remains in liquid form throughout its cycle. To compensate, 1P D2C relies on a much higher pumping power to fulfill the cooling requirements of AI workloads—comparatively, our two-phase technology has a 4-9x lower flow rate.

Single-phase’s higher flow rate causes higher rates of erosion, raising the maintenance costs associated with repairing and replacing critical components. It also inevitably spikes energy costs to maintain this elevated pumping power.

The gulf between 1P and 2P D2C’s flow rates is only set to widen. Already, NVIDIA announced a need to efficiently cool its Vera Rubin supercomputers by significantly raising 1P D2C’s flow. As evolving AI workloads command greater levels of cooling, this trend in flow rate disparity will magnify until two-phase will consume only one-tenth of the energy that single-phase needs to maintain acceptable thermal performance.

Total Harmonic Distortion (THD)

Our lower flow rate also allows us to avoid a lesser-known but impactful cost: Total Harmonic Distortion (or THD). In simple terms, an electric current is meant to flow smoothly in a sine wave. However, THD introduces arrhythmic bumps to these sine waves, in turn “dirtying” the power circulating throughout your facility.

Larger pumps—like those employed by single-phase systems—are a key culprit of THD. To counterbalance THD, facilities or CDU manufacturers need to use harmonic filters that cost roughly $10-15K per unit. From an operating perspective, these filters are ~98% efficient, taxing the CDU’s energy consumption by an additional  ~2%.

The best way to prevent THD—plus the need to mitigate it? Stick with two-phase. Lower flow rates eliminate a need for larger pumps.

Fluid Maintenance

Of course, maintenance costs don’t stop at hardware. With single-phase, you have to maintain the fluid itself.

It’s often said that “water is the source of life.” In water-based systems like single-phase, that “life” takes the form of microbacterial growths that can corrode your tubes and degrade your fluid. If not properly maintained, this biofouling can force your cooling to work below optimal efficiency, forcing operators in turn to compensate by further increasing flow rates—thereby worsening the other problems we’ve outlined above.

What exactly does proper fluid maintenance entail? For starters, regular sampling and lab testing (on a quarterly or monthly basis) to determine rates of biofouling—a process that’s hampered by limited lab availability and lengthy waitlists. Periodic flushing is also required, which means you’ll continually need to buy new single-phase fluid to replace what’s discarded. You’ll also need filters, leak detection sensors to prevent water hitting critical electrical equipment, and a host of other components needed to keep 1P D2C at its best.

Oh, and you’ll need a whole new workforce. Who else is going to do the sampling, flushing, complying, repairing, and maintaining that single-phase demands?

A Whole New Workforce

Water may bring a certain degree of familiarity—but it also brings an entire service economy behind it. Skilled labor and operational overhead are drastically higher for single-phase and its complex service procedures. In an industry already struggling with skilled labor shortages, an endless search for expert technicians will damage single-phase’s scalability.

The industry has already signaled this growing need for proper fluid maintenance professionals. Vertiv’s recent $1B acquisition of PurgeRite, which specializes in “mechanical flushing, purging, and filtration,” indicates that this labor problem is real and recurring. And even if data centers rely on third parties to handle single-phase maintenance, those costs will compound throughout their facility’s lifecycle—and could’ve been entirely avoided if they’d switched to two-phase instead.

No water? No worries.

By now, it should be clear: because of everything we’ve described above (and more), 2P D2C delivers a 44% reduction in CDU + fluid maintenance OpEx vs. single-phase.

Ask yourself: do you know everything it takes to keep your current cooling system running at peak performance? Unlike single-phase, NeuCool^® has a streamlined design that’s purpose-built to minimize your data center’s energy use—and your cognitive load. It’s a “set-it-and-forget-it” solution, empowering you to focus on other matters while it delivers the industry standard of chip-level cooling.