One of the biggest PR challenges currently facing Silicon Valley is the sizable ecological impact of building new AI systems; just this morning, the Wall Street Journal reported that Microsoft has inked a new deal with Chevron to build a sprawling new data center in West Texas powered by natural gas. Now, Nvidia is claiming to have effectively removed at least one major component of the AI industry’s environmental footprint: its dependency on local water supplies.
Traditionally, the bulk of data centers’ water usage comes from keeping the AI chips housed within from overheating as they perform their constant, energy-demanding computations. Nvidia says its breakthrough lies in a new, “closed-loop” cooling system, which recycles a liquid coolant made of three-quarters water and one-quarter propylene glycol—a chemical recipe similar to that used for the antifreeze that goes into a car’s engine.
Crucially, the coolant can remain operational at temperatures of up to 115 degrees Fahrenheit. (For context, that’s significantly hotter than the temperature limit manufacturers build into hot tubs.) That means that Nvidia DSX data centers can avoid overheating without the huge amounts of water and fans that are typically required, all of which adds up to lower operational costs and a reduced ecological footprint.
“Although each generation [of new data center infrastructure] offers significantly more computing power for each watt, full liquid-cooled AI compute infrastructure enables data centers to dramatically reduce cooling energy consumption—making a meaningful difference to overall data center energy use at hyperscale,” Nvidia wrote in a blog post published Monday.
A politically opportune moment
Nvidia’s new cooling system, which it says can produce “up to a 100% reduction in water use,” will definitely catch the attention of the AI industry at a time of growing disenchantment with AI’s ecological impacts among the public.
A recent survey conducted by the Pew Research Center found that the majority of Americans who are familiar with data centers view them as being “mostly bad” for the environment, as well as for household energy costs and the quality of life for people living in neighboring areas. (Some Republican lawmakers have been blaming this wide-scale public resentment towards data centers on alleged misinformation campaigns carried out by leftist environmental activists and the Chinese Communist Party, who are aiming to undermine the American AI industry.)
Picking up on those political headwinds, both Google and Microsoft recently announced new efforts to cut back on the water usage by their data centers, alongside a push into so-called “edge” computing, through which AI tools are powered by local devices rather than data centers.
Even more importantly than their environmental benefits, Nvidia’s new cooling system will appeal to an AI industry that’s becoming increasingly desperate to offer users cheaper access to AI. If tech companies can reduce their overhead costs by cutting back on the power used by their data centers, that could provide them with a route towards cutting back on the price of tokens, which are the basic unit used to measure AI usage.
Water concerns could also help nudge investors and developers toward the idea of building data centers in space, an idea that sounds like it was pulled from a cyberpunk novel but is in fact starting to be taken very seriously within companies like SpaceX and Google. Since data centers could simply release any excess heat they generate into the vacuum of space, they wouldn’t need to rely on fans or an external water supply. Whether or not such facilities are viable in practice, however, is still very much an open question.
Bigger concerns
There are some major caveats to Nvidia’s claim of a 100% water-free cooling system, however.
For one thing, the efficacy of the system varies depending on the climate in which it’s running. If outdoor temperatures approach the 115-degree Fahrenheit mark—as they sometimes do in hotter regions like Arizona and Nevada, where many data centers are being built—then additional cooling resources will be needed. Nvidia acknowledges this in its post: “A data center in the Scottish Highlands and one in Phoenix, Arizona, face very different realities,” the company wrote. “But even in warmer climates, the shift toward [115-degrees Fahrenheit] coolant moves operators significantly closer to that chiller-less ideal—where chillers may turn on just a few days a year when the outside air temperature demands it.”
Additionally, though Nvidia is, far and away, the biggest manufacturer of AI chips, it’s too soon to say whether or not the new design will be scalable across the industry; the company made no mention of cost in its new blog post and didn’t immediately respond to a request for comment on this detail. It’s likely that, at least for the foreseeable future, many of the biggest AI labs will continue using traditional, thirstier data centers.
It’s also important to bear in mind that water usage is just one drop in the bucket of the industry’s total ecological impact. Many of them are powered by fossil fuels, pouring copious amounts of greenhouse gases into the air. The huge amount of investor dollars and political energies that are being funneled towards the construction of new data centers, meanwhile, is coming at the expense of the rollout of more sustainable alternative models, such as wind and hydroelectric power.
Read the full article here
