The artificial intelligence boom isn't just straining electricity grids; it's creating a supply crisis for the industrial equipment that powers data centers. Gas turbine manufacturers like GE Vernova are booked solid for years, with orders stretching into 2031, as Amazon, Google, Microsoft, and other hyperscalers race to secure standalone power sources for their AI infrastructure. This bottleneck reveals a critical truth: the real constraint on AI expansion isn't computing chips or software, but the physical power infrastructure needed to run them.

Inside GE Vernova's largest manufacturing facility in Greenville, South Carolina, engineers and factory workers are working around the clock to meet demand. The company hired 200 workers last year and expects 300 more to start by year's end. These aren't small machines; each turbine stands 31 feet tall, weighs 280 tons, and can power roughly half a million homes. One turbine costs more than $250 million, and prices have skyrocketed 300% over the past three years.

The surge in demand reflects a fundamental shift in how AI companies approach infrastructure. Rather than relying on traditional utility grids, hyperscalers are increasingly deploying standalone power sources to avoid grid bottlenecks and ensure reliable, continuous power for their data centers. Microsoft, for example, just purchased seven GE Vernova turbines to power a data center in Texas that will generate 2.7 gigawatts of electricity, enough to power about 3 million homes. GE Vernova turbines are already online at Elon Musk's xAI Colossus 1 campus in Tennessee, and nearly a gigawatt more are being deployed at OpenAI's Stargate project in Texas.

Why Is Power Infrastructure Becoming the Real Bottleneck?

The infrastructure crisis extends far beyond turbine manufacturing. Utility companies are quoting two-to-four-year wait times just for feasibility studies, and sites without existing transmission connections are often rejected outright. A proposed $12 billion data center in Indiana was rejected in September 2025 over land use, water, and safety concerns. Norway has capped new entrants without existing infrastructure at just 5 megawatts, barely enough for a small operation.

The math is stark: a single ChatGPT query consumes 10 times the energy of a Google search. According to Goldman Sachs Research, global data center power use is on track to jump about 50% by 2027 and could surge up to 165% by decade's end. This explosive growth means companies that already own their power connections hold assets that are extremely difficult to replicate.

AI hyperscalers need a specific combination of factors that has become increasingly rare:

• Megawatt-scale power available immediately: Hyperscalers need 100 megawatts or more online within months, not years.
• Renewable energy sources: Microsoft, Google, and Amazon have pledged 100% renewable power, which is scarce at scale and difficult to secure.
• Low-latency connectivity: AI workloads require fiber infrastructure for massive data transfers, ruling out remote sites.
• Cool climates: Lower ambient temperatures cut cooling costs significantly, giving Nordic sites a competitive advantage.
• Political stability and data regulations: Hyperscalers need stable jurisdictions with data sovereignty protections for sensitive training data.

Almost no location checks every box, and the few that do are already taken or face multi-year waitlists.

How Are Companies Securing Power Infrastructure?

Some companies are taking matters into their own hands by acquiring or controlling power infrastructure directly. Bitzero Holdings, a former Bitcoin mining company, recently signed a binding 15-year lease with OneQode Networks for the entire 110 megawatts at its Namsskogan, Norway data center site, worth approximately $2.6 billion in contracted revenue. This deal transforms Bitzero from a mining operation into a contracted AI infrastructure operator with long-duration recurring revenue.

Bitzero's advantage lies in its ownership of power infrastructure as a licensed grid operator in Norway. The company owns its high-voltage feed lines, connects directly to hydroelectric plants, and operates its own substations. This eliminates the fees, middlemen, and bureaucracy competitors face when working with utilities. Combined with Norway's hydroelectric power, Bitzero's all-in electricity cost, including grid fees and taxes, runs roughly 3 to 4 cents per kilowatt-hour, compared to 8 to 12 cents for traditional operators. Bitzero's power isn't subject to grid curtailment, and its 100% hydroelectric supply means no exposure to gas price spikes or carbon regulations, both critical for AI companies' multi-year commitments.

"Right now, when you need power at scale and you need firm power, the industrial gas turbine is one of the leading solutions for that," said Pablo Koziner, chief commercial and operations officer at GE Vernova.

Pablo Koziner, Chief Commercial and Operations Officer at GE Vernova

The OneQode deal is significant because it demonstrates the economics of AI infrastructure. OneQode will pay for power on top of the lease, run the GPUs, and take the technology risk, while Bitzero collects rent on infrastructure it already owns and powers at industry-low rates. The lease generates implied annual revenue of roughly $178 million at full capacity with an 85% margin. Once the facility commences operations in the first half of 2027, Bitzero's pro forma revenue will jump from roughly $25 million today to approximately $203 million, an eightfold increase.

Are There Alternative Approaches to Data Center Power?

While gas turbines and traditional power infrastructure dominate the current landscape, some companies are exploring unconventional solutions. Underwater data centers, for instance, offer significant cooling and power efficiency advantages by using seawater as a natural heat sink instead of energy-intensive cooling systems. Microsoft's Project Natick demonstrated that underwater facilities could achieve a power usage effectiveness (PUE) of 1.07, compared to the 2024 industry average of 1.56, meaning underwater data centers use far less energy for cooling.

China's HiCloud Technology brought the world's first wind-powered commercial underwater data center into full operation off the coast of Shanghai in 2025. The $233 million facility operates at 24 megawatts and uses at least 30% less electricity than on-land data centers thanks to natural cooling. The facility is connected to a nearby offshore wind farm and supplies 97% of its energy from renewable sources. HiCloud intends to expand the Shanghai facility nearly twentyfold, potentially reaching 500 megawatts of total capacity.

However, underwater data centers face significant challenges. Pods must remain completely welded shut throughout their operational life, making upgrades impossible in a fast-moving technology space. Routine maintenance must be conducted entirely remotely. Additionally, seawater's corrosive nature can degrade efficiency over time, and there are environmental concerns about heat dissipation warming surrounding marine ecosystems. With only one company deploying them at scale, underwater data centers cannot quench the current extreme demand for data center resources.

The power infrastructure crisis is reshaping investment priorities. Smart money is pivoting from tech chips to power infrastructure, recognizing that companies controlling power assets hold the keys to AI expansion. As demand for data center power continues to surge, the ability to secure reliable, renewable, and cost-effective electricity will determine which AI companies can scale and which will face bottlenecks.

" }