From nuclear to quantum computing, how Big Tech intends to power AI’s insatiable thirst for energy

A huge upswing in the number of data centers worldwide shows no signs of slowing down, prompting Big Tech to consider how best to power the artificial intelligence revolution.

Some of the options on the table include a pivot to nuclear, liquid cooling for data centers and quantum computing.

Critics, however, have said that as the pace of efficiency gains in electricity use slows, tech giants should recognize the cost of the generative AI boom across the whole supply chain — and let go of the “move fast and break things” narrative.

“The actual environmental cost is quite hidden at the moment. It is just subsidized by the fact that tech companies need to get a product and a buy-in,” Somya Joshi, head of division: global agendas, climate and systems at the Stockholm Environment Institute (SEI), told CNBC via video call.

A wave of data center investment is expected to accelerate even further in the coming years, according to the International Energy Agency, primarily driven by growing digitalization and the uptake of generative AI.

It is this prospect that has stoked concerns about an electricity demand surge — as well as AI’s often-overlooked but critically important environmental impact.

Data centers, which consume an ever-increasing amount of energy, represent a key piece of infrastructure behind modern-day cloud computing and AI applications.

Giampiero Frisio, president of electrification at Swiss multinational ABB, said the engineering group’s data center business has enjoyed remarkable growth in recent years — with the segment on track to grow by more than 24% in 2024.

Frisio said ABB has been well placed in the AI demand boom to supply mid-sized and big-name industry players with all the components needed to run a data center.

“I think the best way to act now is to increase the energy efficiency. That’s the best way because the technology is there, for example the medium voltage HiPerGuard UPS. You can do it, and you can do it tomorrow morning,” Frisio told CNBC via video call.

The HiPerGuard UPS refers to ABB’s industry-first medium voltage uninterruptable power supply, which it says can provide continuous power to large facilities.

“The second one is to move on the liquid cooling, there is no doubt. Again, this is in the optic of better energy efficiency. Why? Because a single rack, you know the black boxes that look like a wardrobe with all the servers inside, the power density of those is going to be four to six times than before,” Frisio said.

“After that, we are talking about five to 10 years from now, it is the nuclear modular system,” he added.

U.S. tech behemoths Microsoft, Google and Amazon have all secured nuclear energy deals worth billions of dollars in recent months as they seek to bring additional energy capacity online to train and run the massive generative AI models behind today’s applications.

The upsurge of generative AI demand has coincided with a push to find more efficient cooling solutions in data centers, particularly liquid cooling — a process in which water is used to lower the temperatures of servers and other electronic equipment.

French power-equipment maker Schneider Electric recently completed an $850 million deal to take a controlling stake in Motivair Corp, a U.S.-based company that specializes in liquid cooling for high-performance computing.

Schneider Electric CEO’s at the time told CNBC that the all-cash deal, which is designed to bolster its offering to data centers, was “rich, but not overly expensive” and “fits great” with the firm’s strategy.

Alongside nuclear energy and liquid cooling technology, some tech players have suggested developments within AI could help to decarbonize data centers.

Former Google CEO Eric Schmidt, for example, said last month that since “we’re not going to hit the climate goals anyway”, investing in AI could be pivotal to solving some of our biggest environmental challenges.

SEI’s Joshi flatly rejected this point of view.

“These arguments are not new, they are very much in line with the sort of ‘silver bullet’, ‘tech will save us’ rhetoric,” Joshi said.

“There is something inherently at odds with saying we operate within certain finite planetary boundaries and yet by exceeding them and continuing with the same extractive narratives, we are somehow going to solve the problem that we’re in now,” she added.

“I think in the summer of every great technology we discover there is a winter — but don’t pay attention to it until winter arrives,” Raj Hazra, CEO of Quantinuum, the world’s largest integrated quantum computing company, told CNBC via video call.

“That is my way of describing what is happening with generative AI, the infrastructure needed to support it [and] the massive data centers that have to be built.”

Hazra said optimism over the generative AI boom is already straining the cost of running the technology.

“As wonderful as AI is, there are two challenges to it. One is, is it sustainable from a resource standpoint? The other is, is it responsible?” Hazra said. “The reason I bring up this context is because quantum is critical to both of them.”

Quantum computing refers to an area of computer science that uses the laws of quantum mechanics to solve extremely complex problems.

Hazra said a “who’s who” of businesses and strategic investors have recently expressed an interest in Quantinuum, with the company raising around $300 million in its latest equity funding round. Honeywell, which is the firm’s majority shareholder, said the fundraise put Quantinuum’s valuation at $5 billion.

“One of the things that has become quite apparent is it’s no longer OK to say I have a solution to a problem; you have to say I have a sustainable solution to a problem,” Hazra said.

The CEO said one of quantum’s biggest contributions to society can be to make AI both sustainable and responsible.

“I predict that in the next three to five years, you will see people say, what is my compute infrastructure for running my business? It will be a combination of high-performance computing, AI and quantum,” he added.