Could Moving Data Centers Underwater Reduce AI’s Environmental Footprint?

Image

Eric Masanet, a professor and Mellichamp Chair in Sustainability Science for Emerging Technologies at the Bren School, was featured as an expert source in an article from The National Desk examining whether underwater data centers can ease the environmental toll of artificial intelligence (AI).

The story examined underwater data centers as a proposed response to AI's growing electricity demands, with systems cooled by the surrounding water and, in at least one deployment, powered in part by offshore wind. The scale of AI's energy demand makes the potential benefits — and tradeoffs — of the approach worth considering.

Masanet, whose research focuses on energy systems, climate change mitigation, and data centers and information and communications technology (ICT), was asked to weigh in.

Can putting a data center underwater really reduce its energy use and carbon footprint?

Masanet said underwater data centers can reduce the electricity needed for cooling by using "once-through" cooling, a method already common at power plants, in which surrounding water continuously draws heat away from the servers and carries it back into the water body. That reduction in electricity use lowers operational carbon emissions regardless of the power source, though the size of the carbon benefit still depends on how that electricity is generated.

"There may also be a "life cycle" carbon benefit on the equipment side. Given that underwater data centers must be sealed with a controlled atmosphere inside, they might also reduce equipment failures, leading to less hardware waste and fewer replacements over time."

That effect, Masanet added, still needs more data and study.

Does pairing a data center with renewable energy save electricity?

Not directly, Masanet said. Switching to renewable power doesn't reduce the electricity a data center consumes, it only changes the source of that electricity. There are no onsite energy savings from the switch itself, but carbon savings could be substantial if renewable power is replacing fossil fuel-based generation. He added that if the renewable generation is located at the data center, there could be systems-level energy savings too, since electricity loses less to transmission and distribution the shorter the distance it travels.

Could underwater facilities harm marine ecosystems?

Masanet said it's possible, but the field lacks the data to know for sure. Few underwater deployments exist globally, so empirical evidence is limited. On heat, he said hot spots can likely be minimized if the surrounding water body is large enough and mixes and cools naturally, provided sites are chosen and designed responsibly. On maintenance, these systems are generally built to run maintenance-free for long stretches, since servicing them underwater would be technically difficult and expensive. 

Noise and vibration are a bigger open question. 

Masanet said he hasn't seen data on the topic, but pointed to the documented effects of ship noise and vibration on whales as a reason the issue deserves serious research, particularly if underwater deployments start scaling up.

Could underwater data centers scale beyond coastal regions?

Masanet said underwater data centers could technically work in both saltwater and freshwater, as long as the water body is large enough, with sufficient natural mixing and cooling. Whether they should be deployed comes down to weighing benefits such as lower cooling energy demand, potentially longer equipment lifespans, and coupling with renewable sources like offshore wind or wave energy, against costs like thermal pollution, disruption to aquatic environments, and noise. He said more study and data are needed to weigh those tradeoffs, given how new these deployments are. 

At scale, he said, underwater data centers are unlikely to fully replace land-based facilities. 

Terrestrial facilities offer easier access and maintenance, onsite security, tenant access, and lower latency when located near population centers. They also have the advantage of conventional construction and equipment, existing fiber connections, established power grids, and access to land-based energy sources such as geothermal.

Is the net effect on the environment positive or negative?

Masanet said it's too early to tell. He pointed to four reasons why: 

"What we really need," Masanet said, "is a deep scientific study based on actual data."

The full article is available at The National Desk. Masanet's faculty profile and research are available on the Bren School website.