With individuals, companies and governments everywhere realizing the remarkable power of artificial intelligence, a technological arms race has begun, the likes of which the world has not seen since the Cold War. On the front lines is Big Tech, which has been pouring billions of dollars into machine learning ventures in what The Wall Street Journal has called one of the largest investments in U.S. history.
That grand investment promises to reach virtually every corner of the global economy, from healthcare and transportation to banking and retail, to say nothing of the technology’s infinite potential for military and defense.
But AI’s lucrative future is not guaranteed. There’s at least one major hurdle that investors and technologists must overcome in order to turn AI’s potential into reality, and that is algorithms’ unquenchable thirst for electricity.
AI Data Center Energy Consumption: Reaching Capacity
Despite incredible demand, server farms serving AI users have all but reached capacity, with vacancy in North America at a record-low of 1.6%, according to CBRE. As a result, unimaginable resources are being poured into constructing AI data centers and the critical electrical infrastructure needed to support them.
“The AI arms race is essentially becoming who can figure out how to power these data centers the fastest, to get them online, to conduct research and to get the new models developed,” Lee Shaver, a senior energy analyst at the Union of Concerned Scientists, told The Forecast in an interview.
Thanks to AI, global electricity demand is set to grow by more than 1 trillion kilowatt-hours per year through 2030, according to Morgan Stanley. The International Energy Agency says that’s equivalent to Japan’s entire annual electricity consumption. AI’s resource-intensive nature means the technology could account for 20% of the world’s energy usage by the turn of the decade, adds Penn State’s Institute of Energy and the Environment.
“There’s just limited capacity everywhere on the grid to be able to connect these things,” Shaver said. “The growth of data centers is stress-testing all of the systems that are used to plan, permit, build and operate electrical infrastructure.”
Shaped by exorbitant quantities of data, these impressive and rapidly evolving AI algorithms have created an unprecedented surge in global electricity demands. But their continued advance may be hindered by energy shortages as soon as 2027, with some regions already struggling with electrical needs from data centers.
“Right now, the bottleneck is the ability to get energy to run the data centers,” Shaver said. “There are bottlenecks around GPU availability, but that’s an easier hurdle to overcome than the connection of power. If you wanted to buy a new natural gas power plant right now, the main component of that is on a seven-year backorder. So there’s huge delays with that.”
AI workloads are also uniquely tricky for an outdated electrical grid that has been optimized over years for other uses.
“Data centers for AI represent a huge increase in load, a totally different type of load than the market or grid has seen before, and they are being added to the grid much faster than any other type of load,” Shaver said.
“Traditionally, the biggest loads have pretty predictable patterns and synchronize through the grid. When they’re training an AI model, all of the GPUs inside of a data center are essentially operating in sync, which means that there are massive fluctuations that can change in seconds. They can actually damage different components of the grid.”
Rethinking the Cloud
As the electrical grid reaches its physical limits, companies must adapt and adopt a cloud-appropriate model that optimizes loads based on the most reliable power access. This fundamental transformation means IT choices must be assessed carefully based on energy availability, costs, which are becoming intrinsically linked to power, and latency, where computing should be done closest to the source.
“2026 will be the year that businesses move from AI-first to AI-smart,” Nutanix CEO Rajiv Ramaswami recently told IT news outlet CRN as part of its CEO Outlook 2026 project.
Perhaps platform optionality and the ability to move workloads seamlessly across environments, from private to public cloud to the edge, are the most durable foundations of AI. Avoiding vendor lock-in and maintaining flexibility are competitive necessities in the age of machine learning. In fact, Gartner predicts that 40% of leading enterprises will adopt hybrid computing by 2028 as they spread workloads across different architectures. No longer can cloud alone be a default destination, as companies must adapt to the shifting economics of AI as they embark on cloud 3.0.
Winning the AI Arms Race
On the subject of AI data center energy, news outlets have been clear: The stakes are incredibly high, with potential impacts for geopolitics, the global economy and the environment.
In November 2025, Jensen Huang, CEO of Nvidia — now the most valuable company in the world thanks to its role in supercharging AI — told the Financial Times that China is poised to win the AI race due to its increasing electrical production and regulatory flexibility in building out additional infrastructure.
Per TIME magazine, the United States has greater access to advanced chips, and currently has more sophisticated AI models and significantly more data centers overall. China, however, leads the world in AI patents and is producing more than double the amount of electricity than the U.S., according to Forbes. Unsurprisingly, hyperscalers are expected to spend $1 trillion on energy infrastructure over the next year to stay competitive, Morgan Stanley reports.
With the planet already facing monumental impacts from global climate change, a crucial question arises about how future power plants will be operated and whether they will aid or hamper vital environmental efforts.
Powering the Future
“A year ago, new data centers were planning almost exclusively to use natural gas. But we’ve seen that start to change over the last few months, where more and more data centers are committing to building or buying renewables to meet that demand,” Shaver said.
For example, Google recently announced two new colossal data centers in Michigan and Minnesota that will utilize wind, solar and battery technology for cleaner power production.
In the dire race to reliably power an AI-centered future, nuclear energy is re-emerging, as well. In fact, the Three Mile Island nuclear plant, made infamous by a reactor meltdown in 1979, is reopening next year to power Microsoft data centers, according to reporting from NPR.
“The massive, rapid need for power is driving some urgency to conversations about things like energy efficiency, demand flexibility, virtual power plants, etc,” Shaver said. “All of these other solutions to having a more sustainable grid that have been on the back burner for the last few years are getting a lot more interest.”
Other companies are turning away from renewables and even the electrical grid as a whole. Meta, for example, is building an entirely off-grid power plant run by natural gas, The New York Times reported in March 2026.
“Natural gas is a greenhouse gas that traps more of the energy from the sun within our atmosphere, resulting in global warming, leading to obviously higher temperatures, but also more frequent storms and more extreme weather,” Shaver said.
“In addition to the carbon emissions, there are also air pollutants that result in respiratory and cardiovascular challenges.”
Because energy availability is the single biggest obstacle to data center development, all server construction plans must account for their own immense energy needs. If they don’t, they could encounter significant delays, according to Bloom Energy, whose “2026 Data Center Power Report” cites a “power expectation gap” of 1.5 to 2 years for hyperscalers developing new data centers.
Companies and governments are betting unthinkable sums of money on what they hope to be the decisive technology of the future. In IT, the shifting economics of machine learning are triggering a significant change from cloud-native to cloud-smart.
As the sprint for AI supremacy intensifies globally, experts insist that the race will be won and lost by infrastructure. If they’re right, the geopolitical world order might be decided as much by the facilities that power AI as it is by the actual outputs of AI models themselves.
Chase Guttman is a technology writer, an award-winning travel photographer, Emmy-winning drone cinematographer, author, lecturer and instructor. His book, The Handbook of Drone Photography, was one of the first written on the topic and received critical acclaim. Find him at chaseguttman.com or @chaseguttman.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
