How Nvidia’s New Cooling Technology Could Reshape the AI Industry as We Know It

Listen to the audio version of this article (generated by AI).

Nvidia (NVDA) published a seemingly innocuous blog post on its website Sunday morning. But the news could literally transform the entire artificial-intelligence (“AI”) industry.

In the blog post, Nvidia’s head of corporate sustainability, Josh Parker, revealed that Nvidia’s new closed-loop liquid-cooling system may have solved one of AI’s biggest problems by reducing data-center water use to near zero and cutting cooling-related energy consumption by up to 40%.

This breakthrough has the potential to dramatically reshape the economics of AI. By eliminating most chillers and fans, possibly eliminating water for cooling purposes, and reducing energy usage, server racks equipped with Nvidia’s Rubin architecture can help AI data centers operate more efficiently while freeing up space and power for additional compute capacity.

Parker wrote this in his post:

Hot tubs sit at about 38 to 40 degrees Celsius, warm enough that most people can only soak for about 15 minutes. Nvidia’s newest AI servers can run their cooling liquid even hotter – up to 45 degrees Celsius, or 113 degrees Fahrenheit. That higher temperature limit is precisely what makes them more energy efficient.

This is a huge development. By achieving this type of cooling, Nvidia may have solved – or at least gone a long way toward solving – the massive energy and water challenges that plague AI data centers and the communities in which they operate.

How Nvidia’s New Liquid Cooling Works

In Parker’s blog post, he noted that Nvidia’s Rubin generation of AI architecture is the “world’s first to achieve 100% liquid cooling – every chip, every networking component, cooled entirely by liquid in a closed loop with no fans anywhere in the system.”

Here’s how it works.

Nvidia’s liquid coolant is a mixture of 75% water and 25% propylene glycol, a formulation similar to antifreeze. The key is that it can operate at 45 degrees Celsius (113 degrees Fahrenheit). This heat tolerance allows data centers to rely less on – or eliminate – the chilling equipment that’s responsible for much of the energy and water consumption.

That’s because, with Nvidia’s 45-degree liquid cooling, heat is captured directly at the chip and transported through liquid loops operating at much higher temperatures. This allows outdoor dry coolers to reject heat efficiently while substantially reducing the need for chillers, cooling fans, and water consumption.

There’s no loss in processor performance either, because Nvidia’s hot-water cooling keeps device temperatures within optimal operating range, even with the coolant at 45 degrees Celsius.

Importantly, Parker explained that no parts of the server depend on cool air. The liquid performs the cooling within a closed-loop design that is filled with coolant just once and remains closed for as long as the data center is operational.

This cooling architecture, in the most moderate and cool climates, could reduce data-center water usage from roughly 2.6 million gallons per megawatt per year… to nearly zero water used for cooling – a saving of up to 100%.

Ali Heydari, Nvidia’s director of data-center cooling and infrastructure, stated:

With dry-cooler-based designs, it’s a closed-loop system with no evaporative water cooling – outside of maybe 1% of the year when we might need chillers in some [warmer] climates.

This can potentially save hyperscale data centers millions of dollars in water and energy costs per year. Not to mention invaluable natural resources. And that can’t be understated, especially considering the public backlash over the construction of AI data centers in some communities.

Why Cooling Is So Critical in AI Data Centers

Cooling is one of the most important aspects of an AI data center. Without proper cooling, AI components would literally break down and/or melt from the heat they produce. And that would bring a data center to its knees.

My MarketWise colleague Jonathan Rose covered some of the top cooling stocks to own in an April 30 article.

In it, he wrote:

Every AI chip running a workload generates heat. As Nvidia’s flagship Blackwell rack architecture scales toward 100 kilowatts per rack – and as AMD and custom hyperscaler silicon push densities even higher – the heat problem becomes the binding physical constraint on whether a data center can run at all.

Air cooling cannot solve this at scale. The physics of moving enough cool air through a 100-kilowatt rack is a losing battle. Liquid cooling – pulling heat directly off the chip with a fluid loop – is the only proven solution that works for the rack densities AI requires.

Graphics processing units can reach temperatures of 105 degrees Celsius when processing heavy AI workloads. That’s 221 degrees Fahrenheit… in other words, blazing hot.

Cooling Is Essential, but It Comes With a Heavy Environmental Cost

With data centers and their equipment in constant need of cooling, air and water have typically been the primary solutions. The problem here is that large data centers can consume up to 5 million gallons of water every day. That’s roughly the amount of water used by a city of 50,000 people, according to the Lincoln Institute of Land Policy.

This can cause water shortages, especially in areas already prone to droughts. Georgia, for example, has recently experienced a moderate-to-severe drought. Meanwhile, as the state suffered from low water levels, a data-center campus in Fayetteville (around 20 miles outside of Atlanta) managed to drain nearly 30 million gallons of water from the area – without paying for it and without anyone noticing until local residents experienced low water pressure in their homes, Politico reported.

For context, 30 million gallons of water could fill 44 Olympic-sized swimming pools.

But water is only one of the resources data centers devour daily.

Last fall, Corey McLaughlin, editor of the Stansberry Digest newsletter, analyzed how much power American data centers actually need:

The Environmental and Energy Study Institute estimates that U.S. data centers could require as much as 130 gigawatts of power every year by 2030.

That’s enough to power, on average, about 114 million homes for a year. And that’s the equivalent of a huge chunk of all of the housing (148 million units) in the U.S.

Not only would this new demand be a huge draw on utilities, but it would also mean data centers accounting for 12% of all electricity used in the U.S. (versus about 4.4% at the end of 2023).

I also covered the insatiable power needs of AI data centers earlier this year, writing:

A recent study from Gartner projects that electricity demand fueled by AI data centers will double by 2030.

Right now, data centers consume 448 terawatt-hours (“TWh”) of power… That figure is expected to rise to 980 TWh by the end of the decade.

That would represent roughly 3% of the entire world’s power consumption.

For context, 980 TWh could:

• Power a country the size of Germany for a full year
• Provide power to around 93 million average American homes for an entire year
• Give electric vehicles enough juice to drive billions of miles

Closing the circle here, Nvidia’s Parker noted that, according to McKinsey, up to 40% of a data center’s electricity consumption comes from cooling.

Something had to give, from both a resource and cost perspective.

Nvidia seems to have found the solution.

Expect Nvidia’s 45-Degree Liquid Cooling to Be a Game Changer Over Time

Most data centers are loud and have widely varying indoor temperatures. This is the result of traditional air-cooling methods.

Like cooling fans that can reach 85 decibels or louder. And “cold aisles,” where the servers are strategically arranged, which can get as chilly as 64.4 degrees Fahrenheit (data centers are not actually walk-in freezers as some people imagine). And “hot aisles,” where the heat and exhaust are expelled from the servers, which range from 95 degrees to 110 degrees Fahrenheit.

Moving forward, however, every data-center operator and cloud provider that uses Nvidia’s Rubin platform will automatically benefit from the company’s new cooling technology, which is integrated into every Nvidia Rubin architecture.

Not only will Nvidia’s liquid cooling save water and energy, but it will also free up space for more servers and equipment – thus creating more compute capacity per AI factory. Current air-cooling technology occupies a fairly significant amount of real estate. Fully liquid coolers consume considerably less space, with Parker noting that systems previously occupying six rack units now fit in two.

This is a truly exciting development for data-center operators. But there are a couple of caveats.

For one, not every data center will immediately swap out its current architecture for Nvidia’s Rubin platform, which is still in volume production (ahead of schedule, actually) and won’t ship until later this year.

So, we’re talking years before even a small percentage of AI data centers are running on the Rubin platform. Until that happens, data centers will continue to use their current, less-efficient cooling systems.

Also worth noting is that data centers configured for air cooling will need to retrofit their infrastructures to accommodate Nvidia’s liquid cooling.

This will require the installation of direct-to-chip liquid-cooling infrastructure – which involves cold plates installed onto the processors to capture the thermal heat load – to run Nvidia’s Rubin systems. Additionally, advanced cooling-distribution units – which data centers using air cooling do not have in place – are necessary if those data centers want to integrate Rubin hardware.

Data centers will also need heat-rejecting outdoor dry coolers or cooling towers to accommodate the liquid cooling used by Rubin platforms.

All these modifications come at a cost. One estimate puts the price somewhere between $2 million and $7 million per megawatt.

That’s no small price for a data-center operator to pay. And the reconfiguration of server platforms to adapt to liquid cooling is no small feat for hardware manufacturers like Dell Technologies (DELL), Super Micro Computer (SMCI), and Hewlett Packard Enterprise (HPE) either.

But the overall savings may be worth the initial investment in the long run.

As noted earlier, if Nvidia’s hot-water liquid cooling operates as expected, it could cut water consumption to near zero and energy use by up to 40%. That not only equates to several millions of dollars saved for data-center operators, but it also frees up power capacity for more compute.

Essentially, AI data centers will be able to do more for less money, using less power and water. That could revolutionize AI data centers as we know them.

And it’s certainly no surprise that Nvidia – the world’s most valuable company, with a market capitalization nearing $5 trillion  – is the company behind the innovation.

What Nvidia’s Cooling Breakthrough Means for Investors

After closing the week at roughly $210 per share, Nvidia’s stock got a jolt from the liquid-cooling news during the morning of June 22, jumping as high as $213.76 by around 10 a.m. Eastern time. But the pop was short-lived. Nvidia slid below $210 by 11 a.m. and remained there throughout the afternoon.

But, again, this technology will not be implemented in data centers overnight. It will take years before existing AI factories shift to full liquid cooling…if they do at all.

That said, the hot-water cooling solution is just the latest in a long line of innovations from Nvidia. That alone is enough to justify investment in the company. Over the past year, Nvidia’s stock price has gained more than 35%. Go back five years, and it has risen by a staggering 940%.

Nvidia’s hot-water cooling could push the stock even higher in the coming years. The company already has a sizable Rubin backlog.

Hyperscalers like Amazon (AMZN), Microsoft (MSFT), Alphabet’s (GOOGL), Meta Platforms (META), and Oracle (ORCL) have already booked Nvidia’s entire Rubin 2027 production and a good portion of 2028’s for their platforms.

And that is music to Nvidia’s (and investors’) ears.

Regards,

David Engle

Editor’s Note: U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell recently summoned the CEOs of America’s biggest banks to an emergency meeting over one AI lab’s breakthrough. The Fed, Treasury, and virtually every major bank on Wall Street believe it could reshape the entire U.S. financial system. 60-year Wall Street legend and the founder of Chaikin Analytics, Marc Chaikin – whose system pinpointed Nvidia in 2014 before it soared 55,000% – has identified a way to invest in this lab for just $40 before it goes public. But you must act now to take full advantage of this pre-IPO “backdoor.” Click here to learn more.