The race for AI is a race for electricity
The AI buildout has become a race to secure power. Hyperscale campuses that once needed tens of megawatts now need hundreds, and some target a gigawatt or more. Meanwhile, utilities in major data center markets face transmission limits, generation shortages, long interconnection queues and record requests for new load. Power availability increasingly determines where a data center can be built in the first place.
Wind energy is one of the most interesting pieces of that puzzle. It's not because wind can run a data center 24/7 on its own. It can't. It's because wind is cheap, available at huge scale, quick to build compared with most generation, and plentiful in places with open land. Those qualities are pushing developers to rethink how wind and computing fit together.
When we first published this article, it drew on an EnkiAI analysis of major U.S. wind-powered data center projects. Since then, the deals have gotten bigger and the models more creative, even as federal policy has turned against wind.
Why wind can't do it alone
The core issue is that a wind farm's nameplate rating is not the power it delivers. Nameplate is the maximum output in ideal wind. Over a full year, U.S. wind farms produce on average about 34% to 36% of that maximum, which is their capacity factor. Strong sites in places like Texas and the Great Plains with modern, taller turbines can reach 40% to 45% or more.
That means a 200 MW wind farm produces, on average, the energy of a plant running at about 70 to 80 MW around the clock. And it doesn't deliver that evenly. Some hours it produces near full output, and some hours almost nothing.
A data center, by contrast, runs near full load every hour of the year. A multi-billion-dollar GPU cluster can't shut down every time the wind dies. That's why every serious wind-powered data center pairs wind with something else: grid power, battery storage, backup generation or a flexible computing load. You can test the co-location math for any wind farm and data center size with our Wind-to-Data Center Co-Location Sizing Calculator.
Nameplate Isn't the Same as Power You Can Count On
From PPAs to co-location: the deals
For years, the relationship between wind and data centers was simple. A tech company signed a long-term power purchase agreement (PPA) with a wind developer. That gave the developer the revenue certainty to finance a new project, and gave the tech company clean energy credits to match its use. Meta's partnership with Invenergy, including the 196 MW Seaway Wind Energy Center in Texas and about 1.8 GW of clean energy in total, is a classic example.
That model is still growing fast. In September 2026, RWE and Oracle signed a 433 MW virtual PPA for RWE's Panther Creek wind farms in West Texas. Google opened 2026 with 1.17 GW of PPAs with Clearway across Missouri, Texas and West Virginia. Amazon signed with four wind farms in Sweden in August. Hyperscalers are now some of the biggest buyers of wind power in the world.
But a newer model is emerging: putting the data center right next to the wind farm. Soluna's Project Kati in Willacy County, Texas, is a 166 MW data center powered behind the meter by the 273 MW Las Majadas wind farm, co-owned by EDF Renewables and Masdar. The PPA covers the data center's full capacity. The data center can curtail its own operations under certain market conditions and send power back to the grid when demand is high. Soluna's Projects Hedy (120 MW paired with a 200 MW wind farm) and Gladys (150 MW paired with 226 MW of wind) follow the same pattern.
| Deal | Buyer | Developer / Owner | Capacity | Location | Model |
|---|---|---|---|---|---|
| Seaway Wind Energy Center | Meta | Invenergy | 196 MW | Texas | PPA |
| Panther Creek VPPA | Oracle | RWE | 433 MW | West Texas | Virtual PPA |
| Clearway PPAs | Clearway | 1.17 GW | MO, TX, WV | PPA | |
| Sweden wind deals | Amazon | — | — | Sweden | PPA |
| Project Kati | Soluna | EDF Renewables / Masdar | 166 MW DC / 273 MW wind | Willacy County, TX | Co-location (BTM) |
| Project Hedy | Soluna | — | 120 MW DC / 200 MW wind | Texas | Co-location (BTM) |
| Project Gladys | Soluna | — | 150 MW DC / 226 MW wind | Texas | Co-location (BTM) |
Turning stranded wind into compute
Why build a data center in the middle of a wind farm? Because of transmission.
In wind-rich regions like West and South Texas, wind farms often generate more power than the transmission lines can carry to cities. When that happens, the excess is wasted, which is called curtailment, or sold at very low or even negative prices. That stranded energy is a problem for wind farm owners.
A co-located data center turns that problem into an opportunity. It can buy power that would otherwise go to waste, skip some of the transmission bottleneck, and get cheap electricity in return. If its workload can flex, like batch AI jobs, training runs or crypto mining, it can ramp up when wind is abundant and ramp down when the grid needs power. Bitcoin miners have used this playbook for years, and AI developers are now adopting it. Done right, it makes the data center an energy asset for the grid instead of just another load.
Batteries and firm power fill the gaps
Wind and storage work well together. Battery storage can absorb excess wind when it's blowing hard and release it during calm periods or price spikes. A properly sized battery acts as a buffer between variable wind and steady data center demand, and it can earn revenue through energy arbitrage and grid services. You can explore those economics with our Battery Arbitrage Revenue Calculator.
But batteries typically cover hours, not days. Long stretches of low wind, sometimes called wind droughts, still require firm power from the grid, natural gas, or eventually clean firm sources like geothermal, small modular reactors or gas with carbon capture. The strongest strategies combine wind's low-cost energy with storage and firm capacity, rather than forcing wind to do a job it wasn't designed for.
The policy headwinds
Wind's role in AI is growing despite tough federal policy. The 2025 federal budget law phases out wind and solar tax credits for projects that don't start construction or come online within tight deadlines. That's set off a rush to start projects before the window closes. Wood Mackenzie describes U.S. wind growth as a race against the clock, with installations expected to peak around 2027 before uncertainty from tariffs and permitting.
Offshore wind has been hit even harder. Stop-work orders, rescinded port grants, higher domestic content rules and tariffs cut the U.S. offshore wind pipeline roughly in half, from about 56 GW to about 25 GW of planned capacity in a year. For data centers, that means onshore wind in places like Texas, Oklahoma and the Great Plains is the realistic near-term opportunity.
New AI corridors where the wind blows
The bigger story is energy-first development. Instead of hauling ever more power into congested markets like Northern Virginia, developers can put computing where the energy already is. Some of America's best wind regions also have cheap land, room for storage and grid capacity that's otherwise stranded.
That doesn't fit every workload. Latency-sensitive inference may need to stay near cities. But training, batch processing and flexible compute can go where the power is. As AI's power needs grow, some of the strongest wind regions could become the foundation of entirely new AI infrastructure corridors.
Conclusion: a serious piece, not the whole puzzle
Wind won't power AI data centers on its own. Its variability and capacity factor make that impossible without help. But that's the wrong test. The right question is whether wind can be a serious, cost-effective part of the power portfolio, and the evidence says yes.
Hyperscalers are signing bigger wind deals than ever. Co-located projects like Soluna's are turning stranded wind into compute. And the combination of wind, storage, flexible load and firm backup is becoming a practical model for powering AI. Policy headwinds will slow some projects, but cheap, fast-to-build onshore wind remains one of the most important tools developers have. The question isn't wind versus firm power. It's how to combine them.
This article was originally based on EnkiAI's analysis of major U.S. wind-powered data center projects, Meta/Invenergy announcements and Soluna Holdings project announcements. Updated November 2026 with Soluna's Project Kati, RWE–Oracle and Google–Clearway deals, Wood Mackenzie wind outlook and U.S. offshore wind pipeline data.
Chris Kalowes
Founder, WattThe?!
Energy intelligence. Simplified.
