While engineers are ready to build our future, our political leaders are not.
While engineers are ready to build our future, our political leaders are not.
T his month, federal regulators are set to decide how to speed the connection of data centers to the nation’s power grid — a tacit admission that the system is straining under demand the grid was never built to absorb. The fight to come is a fitting symbol of the moment. While Washington has been fighting over tariffs for the past year, a quieter and more immediate threat to American competitiveness has been building in permitting offices and federal courthouses: The infrastructure America needs is not being built fast enough.
The constraint on America’s AI future is physical, not digital. In December, the largest electrical grid in the United States — PJM Interconnection, which serves 65 million people across 13 states — failed for the first time in its history to procure enough power to meet its reliability requirement for 2027. The shortfall was roughly 6.6 gigawatts, about the electricity demand of Philadelphia. Data centers account for 94 percent of the projected load growth driving the gap. Meanwhile, more than 2,000 gigawatts of generation and storage — more than the entire existing U.S. fleet — sit in interconnection queues, with typical waits running three to five years.
We have seen this kind of mismatch before. In the late 1990s, the internet was hailed as a revolutionary technology, and much of the dot-com boom became speculative mania. Firms added “.com” to their names to signal that they belonged to the future. Most of those firms are gone. But the internet itself was not a fad. Between 1995 and 2000, real gross domestic product (GDP) per person grew nearly 18 percent, employers added more than 16 million jobs, and labor productivity rose 17 percent. The IT-producing and IT-using sectors accounted for only about 30 percent of GDP but contributed roughly half of that productivity acceleration, according to work by Dale Jorgenson and co-authors. The bust wiped out investors who confused the technology with the stock prices. It did not wipe out the technology.
The internet did not transform American life by replacing it; it rebuilt what sat underneath. That is the best guide we have to what AI will do. In many ways, ordinary American life looks much as it did in 1995. Families still gather on the couch in the evening to watch television. Commuters still drive to work. Children still go to school, and patients still see doctors. The rituals survived; everything behind them changed. The video store became Netflix. Scheduled programming became streaming. The drive to work — once limited to radio, CDs, or silence, with a phone call from the car difficult for most people to place — now includes calls, navigation apps, podcasts, and audiobooks. The couch is still a couch, and the car is still a car. The economic infrastructure underneath them was rebuilt from the ground up.
AI will almost surely follow the same pattern. Americans will still visit doctors, hire lawyers, and send their children to school. But doctors will use AI to read images and summarize records. Lawyers will use it to review documents that once required armies of associates. Teachers will use it to identify struggling students earlier. Although the fundamental rituals will persist, the underlying infrastructure behind them will change.
But none of this will work without abundant energy. The internet was not built on code alone; it required fiber, servers, semiconductors, data centers, and reliable power. AI will be the same, only more so. The Lawrence Berkeley National Laboratory projects that U.S. data centers will consume between 6.7 percent and 12 percent of all American electricity by 2028, up from 4.4 percent in 2023. AI may feel weightless to the user; it is not weightless to the grid, and it will only continue to grow in the ensuing decades.
This must reframe how Washington thinks about energy policy, much as the 1990s reframed telecommunications policy: Remove barriers, encourage investment, and let the private sector build. A country that wants more data centers cannot simultaneously make it harder to build generation, transmission, and pipelines. Yet new transmission lines routinely take a decade to permit, with federal review under the National Environmental Policy Act and litigation in federal courts adding years of uncertainty.
Congress has discussed permitting reform for years and produced very little. It should try a different mechanism: a bipartisan commission charged with identifying the federal, state, and judicial bottlenecks that prevent timely construction, and delivering a single package on a fast-track up-or-down vote.
The goal is not to weaken environmental safeguards. It is to make review faster, clearer, and more predictable, so projects meeting legitimate standards actually get built. The status quo — endless review, repeated litigation, and a queue measured in decades — protects nothing. It ensures the infrastructure does not exist when it is needed.
AI will produce real prosperity like the internet did, but only if America can build the energy infrastructure behind it. The engineers are ready. Washington is not.