Who Did It First?
Microsoft first came up with this idea back in 2015. Former Navy submariner Sean James basically said; Why don't we just put servers where it's already cold? Next they sank a little test capsule off the California coast called "Leona Philpot" and Leona worked just fine. Then in 2018 they performed a larger proof-of-concept test near the Orkney Islands in Scotland. It ran for two years completely unmanned. Two years underwater and the servers were 8 times more reliable down there than on land. No humidity, no oxygen, no people bumping into things, not data diddling, or physical security breaches.
But Microsoft shelved Project Natick in 2024. They proved the concept would work but they didn't turn it into a real business. China did that. China took Microsoft's research and built an industry. A joint venture between Highlander and HiCloud (Hailanyun) launched the first commercial underwater data center off Hainan in 2023. They have since expanded to a massive 24MW facility off the coast of Shanghai, which became fully operational in this month.
This important distinction matters: Microsoft built the prototype. China built the product.
The Money Part
I'm not a finance person, but I tried to understand the cost difference. Here's what I found.
Building a normal data center costs somewhere between $10 million and $20 million per megawatt. Why so expensive? Land, Concrete, Giant AC units, etc. The Shanghai underwater facility cost $226 million total. At 24 megawatts that's about $9.4 million per megawatt. Because they sank the servers in a steel capsule on the seabed, they did not need land, a building, or AC units.
But the real savings come from running the data center. Traditional data centers are essentially giant ovens. They also drink fresh water (billions of gallons annually) to cool the equipment. Normal data centers use ~ 40% of their electricity just on cooling. Think about that.....almost half the power bill isn't doing any computing, it's just moving heat around. With an underwater data center, you're sitting in a cold bath and the ocean does the work for free (kind of). The Shanghai facility cut total power use by almost 23% and it uses zero fresh water for cooling. Normal data centers drink billions of gallons of water per year from places where water is getting scarce and that's becoming a real problem.
There's a metric called PUE (Power Usage Effectiveness). Perfect score is 1.0. Normal data centers score between 1.4 and 1.6. However, the underwater data center scores below 1.15. That difference translates into real money....millions of dollars a year in electricity savings.
Do You Hear That?
In Vineland, New Jersey, residents near a massive data center under construction report a constant humming that keeps them up at night. One resident, Scott Montgomery, lives about half a mile away and says it sounds like "a huge engine idling all the time." He can hear it through his walls with doors and windows closed. The frustrating part? The facility isn't even fully built yet. As Montgomery put it: "If it sounds like this now, what's it going to be like when everything's up and running?"
It's not just one town—this is happening across the country. Thousands of servers generate intense heat that has to be removed constantly. Large fans and industrial chillers run around the clock, producing noise between 55 and 85 decibels. Many data centers run diesel generators regularly for testing, and some newer AI-focused facilities use natural gas turbines for primary power. These sound like jet engines. For context, some facilities have 356 diesel generators on site.
I know you are going to ask, so the short answer is yes, underwater data centers will affect marine life but the full picture is complex and researchers are still figuring out how serious the impacts might be. I mention "thermal pollution" later, but that's not the only concern. So "impact" doesn't automatically mean "harm" but it does mean change and we don't fully understand what those changes will look like at commercial scale. The honest answer (and the one that makes me uncomfortable) is that we're deploying this technology (on land and under water) faster than we're studying its effects.
What's The Catch?
Well, there are a few things to be considered.
First is the maintenance costs. On land, something fails, someone walks over and swaps it out of the rack. It takes ten minutes. Underwater? If something breaks, you need a ship and a crane. You have to haul the whole capsule up from 115 feet down, break the seal, fix the problem, reseal it and sink it again. That's not a ten-minute job. That's a ten-day job, minimum. Because of this cost, underwater centers rely on extreme redundancy. Microsoft found the failure rate was 1/8th of land-based centers, but when a failure does happen, the repair cost is exponential. This is a bet on reliability over accessibility.
Next, we are putting hot boxes on the ocean floor. Marine ecologists are concerned about thermal pollution. Even a slight rise in local water temperature can drive away native species or attract invasive ones. Companies like Highlander claim independent assessments show the heat is within safe limits. However, scaling this up to 500MW of capacity (which China has planned) creates a heat plume that has never been scientifically modeled at scale. This is a regulatory and reputational risk.
Third, saltwater is brutal. Saltwater destroys metal. To combat this, these capsules are coated with special glass "scales" and anti-corrosive layers. This adds a layer of manufacturing complexity (and cost) that do not effect land-based centers.
The Business Implications
AI is exploding and AI models need massive computing power which means massive heat. Traditional cooling is struggling to keep up. So, what does this mean for business or industries?
- Real Estate: The value of rural land near fiber routes and cheap electricity (like Northern Virginia or Ireland) might decrease. Why pay for acres of land when you can lease a few square meters of seabed?
- ESG (Environmental, Social, Governance): For companies struggling to meet net-zero targets, moving workloads to a facility with a PUE of 1.15 and zero water consumption solves the "water crisis" problem that plagues current AI data centers.
- AI Development: AI training generates a massive amount of dense heat. Traditional cooling struggles to keep up. Underwater centers can handle higher density per square foot because the ocean is an infinite radiator. That being said, we can't ignore the ecological impacts that increased water temperatures will certainly bring.
- National Security: China views this approach as a strategic asset. By co-locating data centers with existing offshore wind farms, they have created autonomous computing hubs that don't rely on a fragile land-based power grid.
My Take
Is the traditional data center dead? No. Not yet. I don't think normal data centers are going away. For stuff that needs low latency like your Netflix loading or a video gaming server you want hardware close to people and probably on land. But for batch processing? AI training? Cloud storage? The math actually makes sense to go underwater. Cheaper to build and cheaper to run. with no water usage. This method is way more energy efficient.
The risks are real: maintenance is difficult and we must control the environmental impacts, but Microsoft proved the concept works and China proved the business works as well.
To me, the future of the data center might just be wet. It is interesting to think about what solutions will be in use by the year 2050 and beyond. Anyway, I am curious what you think. Is this legit or just another tech hype cycle? Anyone actually work in data center operations who wants to tell me why I'm wrong or what I have not considered? Please leave a comment.
Until next time.............
*Thermal pollution - the degradation of water quality by any process that changes ambient water temperature. Thermal pollution is the rise or drop in the temperature of a natural body of water caused by human influence. Thermal pollution, unlike chemical pollution, results in a change in the physical properties of water.