Powering AI with Offshore Wind: The Technology Is Ready. The Policy Isn’t.
China just proved it works. Now comes the harder question.
Every time you send a message to an AI, you use roughly ten times more electricity than a regular internet search. Multiply that by hundreds of millions of people, and you start to understand why the world’s biggest technology companies are in a quiet panic about power. Data centres, the physical buildings that run artificial intelligence, are now projected to more than double their global electricity consumption by 2030, consuming nearly 945 TWh annually, with AI identified as the primary driver. The industry needs land, it needs water, and above all, it needs a lot of electricity, fast. Some engineers think offshore wind might solve all three problems at once. They might be right. But if we are not careful, we will repeat the same mistakes we made on land, just with saltwater involved.
Offshore wind-powered data centres sit at a genuinely interesting intersection of two of the most significant infrastructure stories of our time. The idea is simple in principle: place the data centre close to where the offshore wind energy is generated, use cold seawater for cooling instead of freshwater, and keep the whole operation away from the communities that have been increasingly and loudly saying they do not want this infrastructure near their homes, their farms, or their water supplies. China has already built the first commercial version of this. Western startups are preparing to follow. And the legal and regulatory frameworks that would govern any of it barely exist yet, anywhere in the world. That gap, between what engineers can build and what policymakers have prepared for, is what this blog is really about.
In this post, I will cover what China has already built, what the rest of the world is planning, why communities on land are pushing back so hard, and why, as a wind energy professional, I think the most important conversation right now is not about technology at all.
China Goes First
In late May 2026, Chinese engineers officially switched on the world’s first underwater data centre powered by offshore wind, off the coast of Shanghai in the Lin-gang Special Area. The facility uses seawater as a heat sink instead of freshwater, cuts land use by more than 90% compared to conventional data centres, and eliminates freshwater consumption entirely for cooling. The project is built on straightforward engineering logic: offshore wind generates electricity close to where the infrastructure sits, cold ocean water handles the thermal management for free, and the whole operation stays out of sight of communities who do not want it near their homes.
China is not alone for long. A US startup called Aikido Technologies is developing data centres embedded inside the ballast tanks of floating offshore wind turbines in the North Sea. Each tank can house a 3 to 4 MW data hall, giving each platform a combined compute capacity of 10 to 12 MW. A commercial project off the coast of the United Kingdom may follow in 2028. The CEO of Aikido, Sam Kanner, put the case plainly: “We have this power from the wind. We have free cooling. We think we can be quite cost competitive compared to conventional data-center solutions.” He has also drawn a parallel with the offshore oil and gas industry, arguing that the same logic of moving industrial infrastructure into the deep sea now applies to AI computing. I will come back to why that particular comparison deserves scrutiny.
What Is Happening on Land Explains Everything
To understand why engineers are looking offshore, you need to understand what is happening onshore. Land-based data centres are running into a wall, and it is not a technical one.
A Gallup survey published in May 2026 found that 71% of US adults oppose construction of AI data centres in their local area, including nearly half who strongly oppose them. That makes data centres less popular with neighbouring communities than nuclear power plants. More than $64 billion in projects were delayed or cancelled between May 2024 and March 2025 due to organised opposition, a figure that continued rising sharply through 2025. Between April and June 2025 alone, 20 proposals valued at $98 billion across 11 states were blocked or delayed amid local opposition and state-level pushback, amounting to two thirds of all projects being tracked at the time. Moratorium bills targeting data centre construction are now spreading across multiple US states, with the number of states acting growing steadily through 2026.
This opposition is not irrational. The concerns are concrete. A single large-scale data centre can consume as much electricity as 100,000 homes. In California’s Imperial County, a hyperscale AI facility that would have been the state’s largest was initially approved, then reversed after months of backlash. The developer, having originally pledged to use recycled water, ended up suing the local water authority for 260 million gallons of Colorado River water each year, in one of California’s most water-stressed agricultural regions. The county declared a moratorium. The developer filed a lawsuit to overturn it.
Virginia, the world’s most data-centre-dense region, approved the nation’s first electricity tax on data centres this week, at 1.1 cents per kilowatt-hour, after years of residents bearing rising energy costs and noise from cooling systems and backup generators. Amazon, meanwhile, launched internal investigations into three of its own engineers who testified at a Seattle city council meeting in support of a data centre moratorium, during their off-hours, as private citizens. They were called into HR meetings and told they faced potential disciplinary action. A civil rights complaint has been filed.
University of Michigan researchers, working with the Michigan Environmental Justice Coalition, found that data centres frequently do not deliver the jobs and tax revenue promised to communities, while residents bear rising energy costs and environmental burdens they were given little say over. The grid is stressed too. Research from Texas A&M and Harvard found that training GPT-4 consumed an estimated 50 GWh of electricity, compared to 1.29 GWh for GPT-3, equivalent to nearly 0.1% of New York City’s annual electricity use. The scale of demand is growing faster than the governance frameworks designed to manage it.
Technology Moves Fast. Trust Does Not.
I want to say something that does not get said enough in conversations about AI infrastructure: Artificial Intelligence is genuinely new. Not new in the way smartphones were new, but new in the sense that the people designing it, building it, and running the infrastructure that powers it are themselves uncertain about what it will become and what it will cost. That uncertainty is not confined to the public. It extends to engineers, to researchers, and to the workers inside the companies building this technology who have, in some cases, been willing to risk their jobs to raise concerns about its expansion. Distrust of AI is not irrational. It is, in many ways, the appropriate response to a technology whose long-term footprint, including its energy and environmental footprint, is still being understood in real time. The question is not whether to engage with that distrust, but how to take it seriously rather than dismiss it.
This brings me to the offshore oil and gas parallel that Aikido’s CEO drew. The offshore oil and gas industry did achieve remarkable engineering feats moving infrastructure into the deep sea. It also gave us Deepwater Horizon, decades of regulatory capture, chronic methane leaks, and coastal communities bearing environmental risk while energy profits flowed elsewhere. If that is the model being held up as inspiration for AI data centre expansion, the governance conversation becomes more urgent, not less. The fact that something was done before does not make it a template worth repeating.
Here is what I would add specifically on the marine ecosystem question, because I think it is underappreciated. The ocean is not a separate system operating independently of the rest of the world. It regulates roughly half the oxygen we breathe, moderates global temperatures, and underpins food chains that billions of people depend on. If thermal discharge from undersea data centres creates localised warming that disrupts breeding cycles for marine species, that does not stay a marine problem. It becomes a food security concern, a livelihood concern for coastal communities, and eventually a public health concern. The idea that offshore means consequence-free relies on a view of the ocean as empty industrial space, which it simply is not. We are all downstream of the ocean, even those of us who have never seen it.
My position is this. The offshore route is worth exploring, precisely because it buys time, but only if that time is used to build the governance that the onshore route skipped entirely. A legal scholar at the University of Warsaw recently examined where underwater data centres sit in international law and concluded that existing frameworks covering vessels, sea platforms, and undersea cables do not clearly define what an underwater data centre is, who is responsible if it malfunctions, or what environmental standards it must meet. She found that current international maritime law may not be sufficient and that new regulation may be mandatory. She is right. But she is a legal scholar, not a policymaker. Engineers design the systems. Scientists identify the risks. Neither group has the power to make the decisions that actually determine what happens to the communities and ecosystems affected.
The Finnish model is instructive here. Research on a 21 MW data centre in the Ostrobothnia region found that when a facility was genuinely integrated into local infrastructure, supplying waste heat to district heating networks, it achieved a 36% reduction in district heating CO2 emissions and reduced heating costs by 19%. Local communities were broadly supportive, precisely because the facility delivered real and tangible local benefit. Not a promise of jobs that did not materialise. Not a lawsuit over river water. That is what genuine community integration looks like, and it requires governance before construction, not litigation after.
The sequence I would advocate for is straightforward. Deploy offshore, where the stakes of early-stage failure are lower for nearby human communities while environmental impacts can be monitored at pilot scale. Monitor rigorously, including marine ecosystems, thermal impact, grid integration, and economic viability. Publish the results openly. Build the legal and regulatory frameworks that currently do not exist. Earn the social licence. Then, and only then, come onshore, with evidence, with accountability, and with governance structures that protect communities rather than exploit them. The problem with AI infrastructure is not that engineers lack creativity. It is that the people making siting decisions, technology executives and investors, have almost no accountability to the communities affected by those decisions. Governance is what converts goodwill into binding obligation. Right now, that piece is almost entirely missing, offshore and onshore alike.
So Where Does This Leave Us?
Offshore wind-powered data centres are a genuinely interesting idea whose time may be coming. China has shown it is technically possible. Western startups are preparing to test it at scale. The ocean offers real advantages that land cannot match right now. But the ocean is not empty, and out of sight should not mean out of accountability.
The communities pushing back on land are not anti-technology. They are asking reasonable questions about who bears the costs and who captures the benefits of a buildout reshaping electricity grids, water supplies, and landscapes at extraordinary speed. Those questions deserve answers, not lawsuits, not HR investigations, and not a pivot to the seabed that sidesteps the conversation entirely.
If offshore wind-powered AI infrastructure is going to work, it will be because engineers built it well and policymakers governed it seriously. The first part is underway. The second has barely begun.
Q&A
Doesn’t putting data centres offshore just move the problem? Partly, yes, and that is worth being honest about. Marine ecosystems face real risks from thermal discharge, cable installation, and seabed disruption, and those risks connect to the rest of us through food systems, coastal economies, and global climate regulation. The argument for offshore is not that it is without consequence. It is that the early-stage scale is smaller, the distance from dense human populations creates more room to monitor and learn, and getting it wrong at a pilot scale is recoverable in a way that getting it wrong at commercial scale may not be. That only holds if proper environmental monitoring is mandated from the start, which requires governance that currently does not exist.
Why not just build more renewable-powered land data centres? This is the ideal outcome, and there is already evidence it can work. Research from Finland shows that when a data centre is genuinely integrated into local infrastructure and delivers real community benefit, local support follows. The problem is that the industry’s track record on community engagement has been poor, and trust, once lost, is slow to rebuild. The answer is not to abandon onshore development but to do it differently, with binding commitments, transparent reporting, and accountability structures that existed before the first shovel went in the ground.
As a wind energy professional, are you worried about data centres competing with wind energy for offshore space? Yes. Offshore wind development is already constrained by permitting timelines, grid connection queues, and competing uses of marine space including fishing, shipping, and conservation zones. Adding data centre infrastructure into that picture could either accelerate or complicate development depending entirely on how it is governed. What I do not want to see is the AI industry using offshore wind as a vehicle to bypass the planning and community engagement processes that wind developers have had to navigate for decades. The wind industry earned its social licence the hard way. It should not be handed away to power a technology that has not yet earned its own.
Is offshore wind actually capable of powering large-scale AI computing? At current turbine sizes, a single floating platform with integrated compute can support 12 to 15 MW of data hall capacity. For context, the largest land-based AI data centres operate at hundreds of megawatts. Offshore wind-powered computing is more likely to play a complementary role, handling specific AI workloads or serving as distributed edge computing, rather than replacing centralised facilities entirely. That is an honest description of where the technology sits right now, not a reason to dismiss it.
Did this piece add something to your thinking? You can buy me a coffee with no subscription required, just a small way to support independent writing on energy.
The views expressed here are entirely my own and have no relation to my employer or any organisation I am affiliated with.