NextFin News - The artificial intelligence industry’s insatiable hunger for power has reached a critical inflection point, with Anthropic forecasting that the sector will require at least 50 gigawatts of electric capacity by 2028—nearly five times the peak summer demand of New York City. Speaking at Carnegie Mellon University’s Energy Week on Wednesday, Benjamin Della Rocca, Anthropic’s national security lead, framed this energy surge as a matter of existential importance for both corporate survival and American geopolitical dominance. The projection underscores a widening gap between the Silicon Valley’s computational ambitions and the physical realities of an aging, overburdened U.S. power grid.
The timing of Della Rocca’s appearance was as fraught as his data was stark. He took the stage just days after the U.S. Department of Defense, under the direction of Secretary Pete Hegseth, designated Anthropic an "unacceptable risk" to national security. The Trump administration’s move effectively blacklists the startup from defense supply chains, alleging that the company could theoretically alter its technology to serve private interests over national priorities during a conflict. While Della Rocca declined to comment directly on the legal battle—Anthropic filed two lawsuits against the designation on March 9—he repeatedly emphasized that scaling energy is the prerequisite for maintaining "leadership in artificial intelligence."
The 50-gigawatt requirement represents more than just a massive infrastructure bill; it is a fundamental shift in how the tech industry interacts with the public. Rapid data center development is already driving up electricity prices for residential consumers, a friction point that Anthropic is attempting to mitigate through a new policy announced last month. The company has pledged to cover the costs of electricity price increases resulting from its operations, a strategy that includes investing in curtailment systems to slash power usage during peak demand and potentially funding new generation projects. This move signals a departure from the era of tech companies as mere tenants of the grid to their new role as active, and often disruptive, utility partners.
The search for "the right way" to build this infrastructure, as Della Rocca put it, is leading the industry toward unconventional energy sources. While solar and wind remain staples, the sheer scale and reliability required for frontier model training are pushing firms toward nuclear and geothermal energy. The complexity of this transition is compounded by the need for massive transmission upgrades, a process notoriously bogged down by regulatory hurdles and local opposition. In Pennsylvania alone, planned data centers in Upper Burrell and Springdale are becoming flashpoints for discussions on grid reliability and the economic trade-offs of the AI boom.
The geopolitical stakes are equally high. The Trump administration’s aggressive stance toward Anthropic, contrasted with OpenAI’s recent classified-level agreements with the Pentagon, suggests that energy access and national security clearance are becoming the new "compute" in the AI arms race. If a company cannot secure the gigawatts needed to train the next generation of models, or if it loses the trust of the federal government, its technical prowess becomes a secondary concern. The industry is no longer just competing on algorithms; it is competing on the ability to navigate the volatile intersection of the Department of War and the local utility board.
As the 2028 deadline approaches, the tension between AI’s energy demands and the nation’s infrastructure capacity will likely intensify. The current trajectory suggests a future where the largest AI labs must function as quasi-utilities, managing their own generation and transmission to bypass a grid that cannot keep pace. For Anthropic, the challenge is twofold: it must solve the physical problem of powering its models while simultaneously convincing a skeptical administration that its technology is a strategic asset rather than a liability. The outcome of this struggle will determine whether the next leap in machine intelligence is fueled by American reactors or stalled by an overstretched grid.
Explore more exclusive insights at nextfin.ai.
