Who Powers The AI Revolution—Tech Giants, Utilities Or Both?

Artificial intelligence may run on silicon chips, but its real fuel is electricity. After two decades of steady demand, AI and data centers are causing electricity consumption to soar, which will require utilities and tech giants to collaborate or confront each other. Either way, the aim is for the country to quickly upgrade its network to meet this AI-driven energy surge.

Companies like Meta, Oracle, and OpenAI are building large campuses that require reliable, continuous power. This expansion is testing the capacity of a grid designed for a slower digital economy and prompting utilities, regulators, and tech firms to reconsider their roles, partnerships, and investments.

“Utilities know how to capitalize the cost of building a transformer like the back of their hand, but can’t capitalize a cloud subscription,” Elizabeth Cook of the Association of Edison Illuminating Companies told the audience during a podcast in which we appeared together. That bias toward physical assets has historically limited investment in operations, data analytics, or predictive tools.

Utilities have traditionally been low-risk, capital-heavy institutions. They invest in tangible infrastructure —poles, wires, substations—where costs can be depreciated and returns are assured. The utility industry states that its members often choose to be the first to come in second, implying they let more agile companies lead.

Tech giants can move quickly on high-risk, high-reward projects. Kim Getgen, founder of InnovationForce that hosted the podcast, notes that one hyperscaler can outspend the entire energy sector many times over—by some estimates, as much as fortyfold.

AI mega-data centers like Stargate, backed by Oracle, SoftBank, and OpenAI, are projected to generate $30 billion in annual revenue by 2028. These firms can build data centers in 12–18 months, while new power plants and transmission lines take at least five years to construct and connect. Why’s that?

Utilities make money by persuading commissions to approve capital spending, but operational investments—like grid analytics—generate lower returns. Tech companies, on the other hand, operate in a free market, giving them more flexibility to quickly allocate capital to meet fast-growing demand.

AI’s Soaring Power Demand

According to the International Energy Agency, data center electricity demand worldwide will increase by 130% by 2030. The Department of Energy’s Lawrence Berkeley National Laboratory stated that data centers used about 4.4% of total U.S. electricity in 2023 and, depending on the growth of the rest of the economy, are projected to use between 6.7% and 12% of total U.S. electricity by 2028. It cautions that this depends heavily on AI adoption rates and efficiency gains.

The U.S. faces an unprecedented challenge in expanding its grid. Jeff Weiss, executive chairman of Distributed Sun, explained during a virtual press event hosted by the United States Energy Association: “We need to triple the grid. Everything we do today takes 10 years. We need to figure out how to do it in two.” In practice, that means tripling capacity—not literally rebuilding three new grids. One-third of that new capacity will be needed just to support data centers.

Existing generation, transmission, and workforce limits create bottlenecks. Supply chains for turbines, transformers, and other essential parts are insufficient for quick expansion. Regulatory permitting and interconnection procedures, designed for slower growth, cause further delays. High-powered transmission lines, which span multiple states, are extremely difficult to construct.

Despite having speed and capital advantages, tech giants cannot simply replace utilities. They must follow the same permitting, siting, and interconnection rules. So, it doesn’t matter if you are Google or the hometown utility. Meeting with stakeholders and complying with regulators is part of the process.

“You have to deal with the same federal laws, the same citing, the same public support, and the same supply chain,” says Tom Falcone, president of the Large Public Power Council, during the USEA event. “We deal with these issues in our social construct, in the laws and regulations that we have, and we all have to comply with them.”

However, the problem persists: data centers eager to run advanced AI models urgently need power. Utilities, limited by permitting timelines, supply chain issues, and workforce shortages, rarely meet that urgency. Derek Bentley, partner at Solomon Partners, highlighted the gap at the press event: “You can build a data center in 12 to 18 months. But a new power plant takes five years, plus years more to connect.”

Hybrid Solutions and Partnerships

This isn’t just inconvenient; it’s a fundamental mismatch between the 21st-century digital economy and a 20th-century grid. Progressive utilities are, therefore, working with data centers to develop hybrid solutions.

Indeed, some data centers are colocating with natural gas or nuclear facilities, sometimes combined with renewables and battery storage to enhance scale and reliability. These behind-the-meter setups—where power is generated on-site rather than solely from the grid—are becoming more common. These partnerships enable data centers to access power quickly while maintaining grid stability.

That relieves the burden on the central network, which lowers the risk of blackouts and congestion. Still, for those data centers connected to the main grid—in front of the meter—it results in higher revenues for utilities.

“Hyperscalers are more agile than many utilities, and they are more entrepreneurial and have the capital,” says Clinton Vince, head of the U.S. energy practice at the Denton law firm, during the USEA event. “I do think utilities have been working very well with hyperscalers, although the slower utilities will be disadvantaged tremendously.”

He highlights Meta and Entergy, which are partnering in Louisiana to build major infrastructure supporting Meta’s largest and newest data center, called Hyperion. It will be powered by both fossil fuels and renewable energy.

However, the main criticism is that the regulatory system encourages stagnation. Bud Albright, senior adviser at the National AI Association, pointed out during the press event that “The regulatory format is inadequate today to do the kind of build-out that we need, whether it’s behind the meter or in front of the meter.”

Beyond formal oversight, he adds that public opinion also plays a role. Communities wary of new data centers and transmission lines must understand the broader economic and technological benefits of these projects, from jobs to national competitiveness. Utilities and tech companies must prioritize education and outreach to demonstrate the benefits of their services.

Regulation and Public Perception

Rate design is also under review. Pacific Gas & Electric, for example, requires data centers to pay initial interconnection costs and recover them later as the facility earns revenue. This method ensures costs are shared fairly and stops residential customers from subsidizing commercial loads.

“Affordability is top of mind for us,” says Karen Omelas, director of large load program management for Pacific Gas & Electric. “But we see data centers as beneficial load.”

The energy mix is shifting. Solar and battery storage are expanding rapidly. In fact, storage is becoming a crucial tool to meet peak demand, helping utilities and hyperscalers manage load efficiently. Still, reliable, dispatchable power remains essential. Natural gas fulfills much of this need, while coal is dirtier, expensive, and increasingly irrelevant for electricity production.

All of this highlights a bigger truth: the ongoing transformation is a once-in-a-lifetime event. It’s not about who has the largest balance sheet or who is the most nimble. It’s about collaboration at scale. Utilities need to adopt new tools, rethink their operational models, and partner with the very tech companies they might have once seen as rivals. Meanwhile, tech firms must accept that building computers is one thing; powering them responsibly, reliably, and safely for the masses is another.

For policymakers, the challenge remains just as urgent. Innovation in regulation, faster permitting processes, and public education are essential to prevent congestion that could impede the digital economy. Without action, the infrastructure supporting AI—and the industries it drives—are at risk. And it won’t be because of a lack of creativity or ambition, but because of slow, outdated rules.

The AI revolution isn’t just about chips or advances in machine learning. It’s about wires and power plants. It’s also about the invisible links connecting millions of servers to millions of homes. AI’s ultimate limitation is less about computer intelligence and much more about how fast the grid can expand.