The race to power artificial intelligence is hitting an unexpected roadblock: not a shortage of electricity, but a shortage of the equipment needed to deliver it. Researchers at Johns Hopkins University's Ralph O'Connor Sustainable Energy Institute have identified a critical vulnerability in America's energy infrastructure that could slow AI deployment faster than any power plant constraint. By 2030, demand for transformers, inverters, and other grid-supporting equipment could exceed available manufacturing capacity by roughly 30%, according to their new study and policy brief.

This finding reframes the entire conversation around AI's energy demands. While much attention has focused on whether the nation can generate enough electricity to power data centers, a more immediate crisis may be lurking in plain sight: the large, customized pieces of equipment that actually move that power from generation sites to where it's needed. Unlike modern technologies that can be quickly manufactured or scaled, transformers are labor-intensive, material-heavy products with long lead times and limited production capacity.

Why Are Transformers Becoming an AI Bottleneck?

Transformers are unglamorous but essential pieces of infrastructure. They convert high-voltage electricity suitable for long-distance transmission into lower voltages safe for homes and businesses. For more than 140 years, since William Stanley first demonstrated the technology with George Westinghouse, transformers have been the backbone of the modern electric grid. Yet despite their critical importance, the U.S. manufacturing capacity for these devices has not kept pace with demand.

The challenge is compounded by several factors. First, transformers are customized equipment, not off-the-shelf products. Each one requires significant amounts of labor, materials, and specialized manufacturing capacity. Second, comprehensive national registries do not exist to track U.S. production capacity, making it difficult for policymakers to understand the true scope of the problem. Third, as electricity demand grows from AI data centers, existing transformers will experience greater wear and may need replacing sooner, creating a double squeeze on supply chains already strained by growing demand for copper, steel, and nickel.

The Johns Hopkins team developed an innovative approach to estimate future equipment needs. Since manufacturers keep production information proprietary, the researchers used material demand as a proxy for manufacturing capacity. By understanding the materials that flow into transformer and power electronics industries, they could estimate how many units could realistically be produced. Their analysis linked projected electricity demand from future AI data centers to the specific equipment needed to connect and operate them.

"There is a lot of attention on generation and transmission, and rightly so, but if the transformers aren't there, none of it matters," said Yury Dvorkin, associate professor of electrical and computer engineering and civil and systems engineering at Johns Hopkins. "People think that if they can identify a problem, they can then order a part and make their system work, but there isn't an endless supply of this equipment."

Yury Dvorkin, Associate Professor of Electrical and Computer Engineering and Civil and Systems Engineering, Johns Hopkins University

How Can Policymakers and Industry Address the Equipment Shortage?

The good news is that this bottleneck, while serious, is not inevitable. The research has arrived at a moment when both political parties have focused on strengthening domestic manufacturing and accelerating infrastructure development. Both current and previous administrations have supported production of critical grid equipment, signaling bipartisan recognition that infrastructure speed matters.

• Invest in Manufacturing Capacity Now: The window to expand transformer and power electronics production is narrow. Decisions made today will determine whether manufacturing capacity exists years from now when equipment is needed. Early investment in domestic production facilities can prevent supply chain crises before they materialize.
• Strengthen Supply Chains for Raw Materials: Transformers depend on copper, steel, and nickel. Securing stable supplies of these materials and reducing dependence on vulnerable supply chains will be essential as demand accelerates. This includes both domestic mining and recycling initiatives.
• Coordinate Between Utilities and Data Center Developers: Utilities, data center operators, and equipment manufacturers need better coordination to forecast demand and plan production. Transparent communication about future electricity needs can help manufacturers plan capacity expansions more effectively.

"Everybody agrees we need to build infrastructure faster. What makes this research valuable is that it helps identify bottlenecks before they become crises. The decisions to address them need to happen years before the equipment is needed," noted Abe Silverman, assistant research scholar at the Ralph O'Connor Sustainable Energy Institute.

Abe Silverman, Assistant Research Scholar, Ralph O'Connor Sustainable Energy Institute

A Global Opportunity for American Leadership?

The Johns Hopkins researchers see this challenge as more than a constraint; they view it as an opportunity for American leadership. As countries worldwide pursue AI development and electrification, global demand for transformers and grid equipment is expected to grow significantly. If the United States invests now in manufacturing capacity and supply chains, it could strengthen its own energy infrastructure while becoming a key supplier to allied nations.

This mirrors a historical moment. The first transformer demonstration that helped create the modern electric grid was performed in the United States. The nation has an opportunity to build on that legacy by expanding manufacturing and addressing current bottlenecks. Success could help the U.S. meet its own AI infrastructure needs while supporting partners around the world.

What Does This Mean for AI's Sustainability?

Beyond the equipment shortage, researchers are also working to ensure that AI systems themselves operate as efficiently as possible. A new framework developed by researchers at the University of Sharjah integrates AI-driven renewable energy forecasting with grid economics and lifecycle assessment of AI energy consumption. Their analysis demonstrates that AI-based forecasting can reduce prediction errors by nearly 50% and lower total operational costs by 18.7% compared to conventional approaches.

Importantly, when researchers explicitly accounted for the computational energy footprint of AI systems, they found that the overhead remains marginal relative to the savings achieved. This suggests that well-designed AI systems can actually improve grid efficiency even when accounting for their own power consumption. However, this requires intentional design choices focused on energy efficiency, not just raw performance.

The broader lesson is clear: AI's energy future depends on multiple layers of infrastructure working in concert. Power generation matters. Grid equipment matters. And the efficiency of AI systems themselves matters. Overlooking any one of these layers could create bottlenecks that slow technological progress. The transformer shortage identified by Johns Hopkins researchers is a wake-up call that infrastructure planning must begin immediately, with decisions made years before the equipment is actually needed.