Underwater data centers powered by wind turbines

AI needs power.

A lot of it.

Underwater data centers could help solve the growing electricity shortage facing artificial intelligence infrastructure.

That’s why startup Aikido plans to drop an AI data center on the ocean floor beneath offshore wind turbines. The first step is a small 100-kilowatt pod off Norway this year. If that pilot works, the company wants to scale the idea fast. By 2028 it’s aiming to deploy a 10 to 12 megawatt underwater data center off the United Kingdom.

This is a direct attempt to solve one of the biggest bottlenecks slowing AI growth today.

AI infrastructure is running into a limit that has nothing to do with software. It’s electricity. Training models and running them at scale consumes enormous power. Data center demand is rising far faster than local grids can expand. In many regions, companies already face waits of five to seven years just to secure a new grid connection.

That pressure is pushing companies to explore unconventional locations for compute infrastructure, including space-based data centers powered by solar energy that could eventually run outside Earth’s power grid entirely.

Aikido’s solution is straightforward. Put the data center right next to the power source.

The ocean helps solve several engineering challenges at once. Offshore winds are steadier than winds on land, which means more consistent power output. Cold seawater can cool servers without massive energy-hungry cooling systems. And placing the hardware offshore sidesteps the political fights that often block new data centers near cities.

There’s also precedent that suggests this could work. Microsoft ran an underwater data center near Scotland for 25 months starting in 2018. Out of more than 850 servers, only six failed.

In other words, the technology already works.

The disruption behind the news: Power availability is becoming the main limit on AI expansion.

Building near power generation instead of near cities flips the data center model.

Offshore wind could become part of AI infrastructure.

The traditional model places data centers close to users and then competes for grid capacity.

That model is starting to break. Modern AI clusters can demand hundreds of megawatts of electricity. In some cases they use more power than entire towns. Local utilities can’t expand fast enough to support that demand.

Putting compute at sea avoids that problem. A single large offshore turbine already produces about 15 to 18 megawatts of power. Pairing it directly with a 10 to 12 megawatt data center removes the need for long transmission lines, new substations, and years of local grid approvals.

That changes the economics quickly.

Wind developers have spent the last decade trying to secure power purchase agreements and grid connections. If AI companies start buying power directly at sea, wind farms gain a built-in customer. The turbine becomes both the generator and the landlord.

Cooling is another advantage. Data centers spend a large share of their energy budget removing heat from servers. Submerged pods surrounded by cold seawater reduce that cost significantly. Lower cooling demand means more computing power per megawatt of electricity.

There is also another economic factor. Offshore wind farms often have to curtail power when transmission capacity to shore is limited or delayed. In some European markets, curtailed wind can reach several percent of total generation each year. That means turbines sometimes produce electricity that cannot be sent anywhere.

A colocated 10 to 12 MW data center can use that stranded energy directly at the site. If even 5% of a 1 gigawatt offshore wind farm’s output would otherwise be curtailed, that is about 50 MW of electricity that could run AI compute without building a single kilometer of new transmission.

Then there is politics. Data centers are becoming unpopular neighbors because of their power use, noise, and water consumption. Offshore pods remove most of that conflict. No backyard means no local opposition.

This is why the idea matters. AI demand is exploding while grid expansion moves slowly. Offshore compute offers a workaround.

Part of the reason the energy problem is accelerating so quickly is that AI capabilities have improved dramatically in just the past year, driving a surge in model training and inference workloads that require ever-larger computing clusters.

What to watch next

The next battleground for AI won’t be chips. But power locations.

Wind developers and AI companies are likely to form partnerships.

Expect offshore compute pilots to appear in many places within two years.

The key metric is deployment scale. A 100 kilowatt demo proves the engineering works. A 10 to 12 megawatt installation proves the economics work. If that step succeeds by 2028, the model could spread quickly.

Watch who partners first. Hyperscalers like Microsoft, Google, and Amazon are already searching for power wherever they can find it. Offshore wind operators need guaranteed buyers for their electricity. The incentives line up closely.

Another trigger could be AI training clusters reaching gigawatt scale. Once that happens, building near existing power grids gets a lot harder.

Compute will start moving to wherever energy is easiest to produce.

One of those places is offshore wind.