Post Content
The power sector is grappling with a fundamental mismatch: hyperscale data centers demand electricity at unprecedented speed and scale, while the infrastructure to serve them operates on timelines measured in years, not months. According to Stephen Empedocles, PhD, founder and CEO of Clark Street Associates (CSA), an advisory firm specializing in government funding for technology companies, the industry may be getting some of its assumptions wrong—and the consequences could ripple through grid planning, supply chains, and even geopolitical competition. Empedocles brings more than 25 years of experience as a business development executive in advanced technology across the semiconductor and energy industries. One of the first students of Nobel Prize winner Moungi Bawendi at the Massachusetts Institute of Technology (MIT), Empedocles holds more than 100 patents and was recognized by MIT Technology Review as one of the world’s top 100 young innovators. POWER recently spoke with Empedocles about the gap between expectations and reality for data center power.
One of the most overhyped narratives in the data center space, Empedocles argues, is the notion that these facilities will function as flexible grid assets. The assumption that demand response, virtual power plants, on-site generation, or batteries will enable data centers to contribute meaningfully to grid stability is, in his view, fundamentally flawed at hyperscale. “Downtime carries extremely high financial and operational costs, which limit how much demand they can realistically curtail or shift,” Empedocles explained. “Even with batteries, delivering tens of megawatts of sustained, dispatchable flexibility is challenging, and these systems are typically designed for resilience and uptime, rather than providing ongoing grid support.” What’s underestimated, he said, is how much firm generation and transmission capacity is actually required to serve these facilities. “Data centers will take far more from the grid than they will ever give back,” said Empedocles. “Planning for them as flexible resources risks underbuilding the infrastructure needed to support the incoming scale of load.”
Utilities find themselves caught between the unprecedented urgency of hyperscaler demand and legacy systems that were never designed to match that pace. Without aggressive upgrades to transmission and distribution (T&D) networks, streamlined permitting for power plants, and better demand management, Empedocles warned that the U.S. risks falling behind—not just on infrastructure, but on power availability itself. “Leadership in AI [artificial intelligence] is now determined by who can reliably power data centers at scale, rather than just how quickly they can be built,” he said. “China is outcompeting the U.S. as it moves more aggressively on both generation and grid capacity, and this pace is not something the U.S. can simply catch up to by accelerating traditional infrastructure timelines.” This reality is shaping how utilities approach near-term solutions, including an increased reliance on upgrading existing infrastructure rather than building new. Advanced conductor technologies, for instance, can expand transmission capacity within months compared to the years required for new corridors. At the same time, Empedocles emphasized the need for technologies that reduce the power intensity of data centers themselves, easing pressure on the grid while infrastructure catches up.
Looking ahead, Empedocles expects large, centralized grid infrastructure to dominate federal funding in 2026. He said the next stage of grid modernization support will lean toward large transmission and centralized generation projects rather than distributed energy resource–focused, grid-edge solutions. “Massive AI-driven data centers, along with the dedicated power plants being built to serve them, create a clear federal mandate centered on cost, reliability, and the efficient movement of bulk power,” Empedocles explained. While demand response and virtual power plant models can help address peak conditions, they don’t eliminate the need for new baseload generation, he said. As a result, investments that expand transmission capacity and enable new centralized generation are more likely to move forward than grid-edge concepts predicated on hyperscale loads behaving as flexible resources.
Federal attention on critical minerals should expand beyond rare earth elements, Empedocles argued, to include other high-risk materials such as antimony, gallium, germanium, rhodium, and tungsten. While these materials haven’t attracted the same level of attention, they represent some of the nation’s most vulnerable supply chains. China, Tajikistan, and Russia are the primary sources of U.S. antimony, he noted. “Many of these inputs are critical to power infrastructure, electronics, and industrial systems, but remain vulnerable due to concentrated global control,” he said. The more fundamental issue, however, is processing capacity. Extraction alone doesn’t create supply chain security—federal attention needs to include processing technologies that break from century-old methods. Traditional smelting and refining facilities take years to permit and construct, and require enormous amounts of electricity at precisely the time when AI and data centers are already straining supply. Without processing technologies that materially reduce power intensity and capital cost, expanding domestic processing at scale will remain difficult even with substantial federal investment.
Perhaps the biggest mistake the power sector is making, according to Empedocles, is approaching Washington the same way it did a few years ago. The era of fixed programs with defined applications and expected outcomes is over, he said. “The current administration is aggressively breaking the mold of how government engages with industry and how these deals are structured,” Empedocles said. Federal support is increasingly structured as a bundle of levers rather than a single source. Equity, loans, permitting coordination, and more—these non-cash elements can be what ultimately move a project along. Companies that continue to seek funding through traditional channels are likely to encounter roadblocks or leave value on the table. The organizations that successfully receive support, Empedocles predicted, will be those that come to Washington knowing what the right outcome looks like for their business, then develop a strategy that leverages the full range of tools government can deploy. —Aaron Larson is POWER’s executive editor.
Terms and Privacy Policy