For the first time in nearly two decades, America's electricity demand is rising again, and artificial intelligence (AI) data centers are the primary driver. After remaining essentially flat from 2005 to 2020, U.S. power consumption hit record highs in 2025 and is expected to continue climbing through 2026 and 2027. This shift has profound implications for how the nation powers its digital infrastructure and which energy sources will dominate the coming decades.

What Changed to Break 15 Years of Flat Electricity Demand?

The answer is straightforward: AI adoption. Data centers dedicated to training and running large language models (LLMs) and other AI systems consume enormous amounts of electricity. These facilities run continuously, processing vast amounts of data and performing complex mathematical operations on specialized hardware like graphics processing units (GPUs). Unlike traditional office buildings or factories, AI data centers operate at maximum capacity around the clock, creating an unprecedented and sustained demand for power.

The commercial sector is leading this surge. According to recent reporting, electricity demand growth is being driven primarily by data centers, and the commercial sector is expected to outpace residential demand in 2026 for the first time on record. This represents a fundamental shift in how America's power grid is being used.

Is Power the Real Bottleneck Holding Back AI Progress?

Wall Street analysts and energy experts increasingly believe the answer is yes. While venture capital and corporate investment in AI remain abundant, the physical infrastructure to power these systems is lagging. Goldman Sachs Group released a report concluding that power availability, not capital, represents the most pressing constraint on AI expansion.

"A lack of capital is not the most pressing bottleneck for AI progress; it's the power needed to fuel it," the Goldman Sachs report stated.

Goldman Sachs Group, Financial Analysis

This insight has major consequences. It means that companies racing to build AI infrastructure cannot simply throw more money at the problem. They must secure reliable, long-term sources of electricity. The report suggests that in the short term, hyperscalers and data center operators will rely on a mix of natural gas, renewable energy sources, and "behind-the-meter" solutions (meaning power generated onsite). But for lasting solutions, the focus is turning to nuclear energy.

How Are Companies Planning to Meet This Power Demand?

The energy strategy for AI data centers is evolving into two distinct approaches, each with different implications for how power will be delivered:

• Grid-Connected Model: Data centers connect to the existing electricity grid through utility partnerships, relying on centralized power generation and distribution infrastructure to supply their energy needs.
• Onsite Generation Model: Data center operators build their own power plants directly adjacent to their facilities, producing electricity locally and reducing dependence on grid connections.
• Hybrid Approach: Companies use a combination of grid power, renewable energy, natural gas, and onsite generation to diversify their energy sources and ensure reliability.

Small modular reactors (SMRs) are emerging as a key technology in this transition. SMRs are miniature nuclear power plants that offer several advantages over conventional large-scale reactors. According to Bank of America analysis, SMRs provide lower upfront costs, enhanced safety features, modular design allowing capacity expansion, smaller physical footprints, and reduced carbon dioxide emissions compared to traditional nuclear facilities.

Two publicly traded companies are positioning themselves as pure-play SMR investments: NuScale Power and Oklo. NuScale focuses on utility-scale deployments through direct partnerships with major utility companies, betting that most AI data centers will connect to the grid conventionally. Oklo takes a more direct approach, selling systems directly to data center operators for onsite installation, allowing companies to produce their own power without relying on grid connections.

What Does This Mean for the Future of Energy Infrastructure?

The surge in AI-driven electricity demand is reshaping investment priorities and policy discussions around energy. Nuclear energy, which had fallen out of favor in many regions, is experiencing renewed interest from both the private sector and financial analysts. The Goldman Sachs report emphasizes that meeting AI's power needs will require an "all-in" approach over the next five years, combining multiple energy sources while simultaneously investing in long-term nuclear solutions.

This transition also highlights a critical tension: the infrastructure to support AI's explosive growth cannot be built overnight. Nuclear plants take years to construct, and grid upgrades require substantial planning and investment. Meanwhile, data center operators need power now. This mismatch between immediate demand and long-term supply capacity is driving companies to explore every available option, from renewable energy partnerships to natural gas facilities to behind-the-meter solutions.

The outcome of this energy race will determine not just how quickly AI can scale, but also which regions and countries can attract the most advanced AI infrastructure. Nations and regions with abundant, reliable, and affordable electricity will have a competitive advantage in hosting data centers and attracting tech investment. Conversely, areas facing power constraints may struggle to participate in the AI economy.

For investors, policymakers, and technology leaders, the message is clear: the AI revolution is no longer primarily constrained by computing power or capital. It is constrained by kilowatts. How quickly the energy sector can respond to this unprecedented demand will shape the trajectory of artificial intelligence development for years to come.