The cheapest fix for AI's power crunch may be the grid we already have

The idea: use what we already paid for

As qz.com reported in episode four of its Power Play series, the case for a cheap fix starts with a simple observation. Most hours of most days, the U.S. grid runs at about half its capacity. The wires and poles were paid for decades ago, yet they sit partly idle, partly because nobody agrees on who gets access to them.

Ian Magruder runs the Utilize Coalition, a policy group pushing to get more out of existing lines. He cites a report the group produced with the Brattle consulting firm. It finds that raising grid utilization by 10% over the next decade could save Americans more than $100 billion.

His argument is that the country cannot build power plants fast enough. Getting more from the infrastructure already in place is therefore a faster route to new power. He calls it a "no-regret" strategy, because it avoids the risk of building expensive assets that later sit unused.

Software and hardware fixes for the wires

Paul Denholm, a senior research fellow at the National Renewable Energy Lab, walked through the options. The cheapest are called grid-enhancing technologies. The first is dynamic line ratings, which means tracking local weather to work out how much current a line can carry. Hot wires sag, so cooler conditions leave room for more power. It needs little hardware and sometimes none. Denholm called it "cheap, easy, just do it."

He added a warning. On hot summer days, when demand often peaks, the method helps very little. So the hype needs tempering.

The second is topology optimization, which means switching off parts of the network on purpose. It sounds backwards, but electricity cannot be steered directly, and it tends to flood small lines that might fail. Turning some of those off can raise the capacity of the system as a whole.

Then come the hardware options. Reconductoring replaces old wire, a design about 100 years old, with newer types that carry more current. Devices called flexible AC transmission systems steer power away from lines that cannot cope toward ones that can. Combined, Denholm said, these could substantially raise transmission capacity.

Virtual power plants and flexible demand

The other half of the story is demand. A virtual power plant is a bundle of controllable devices, such as thermostats, electric vehicles, home batteries and commercial loads. An intermediary called an aggregator combines them into a single signal. A grid operator used to dealing with plants of hundreds of megawatts can then treat the bundle as something like a 100 megawatt plant.

This falls under demand response, which means paying people or cutting their bills to use power at different times. Denholm stressed that it is not about leaving anyone in the dark. Typical examples are delaying EV charging by an hour or two, or changing when a dishwasher starts.

Lauren Shwisberg of Rocky Mountain Institute said the technology is mature enough. What holds it back is data access, how utilities are paid and structured, and policy. She noted that gas turbines are backlogged through 2028 and renewable projects wait about five years in the queue. Virtual power plant programs in Ontario and California, by contrast, were deployed in six months to a year.

The investors' view and the limits

Caie Kelley, a partner at the climate venture firm Lower Carbon Capital, represents the money side. One company in her firm's portfolio, Antora, makes heat storage. The episode describes a five gigawatt-hour heat battery announced for South Dakota with partner POET. Kelley argues that pairing solar with batteries and backup power can be the cheapest way to bring a data center online quickly. She also points to the benefit of generating power at home rather than depending on global supply chains. The host disclosed that he is an investor in two of her firm's funds.

Kelley also says demand growth did not begin with AI. Before this wave, her firm expected energy demand in the 2030s to be about double the 2025 level, from other technologies and general growth.

Denholm was the most cautious voice. He said these tools are "definitely meaningful" and well beyond a one or two percent fix, but unlikely to be the entire solution. The reason is that no one knows how much load growth is coming, or how far these mature technologies can be pushed.

Where policy comes in

Magruder pointed to a new Virginia law that passed with bipartisan support. It requires utilities to give their regulator grid utilization figures down to the substation and feeder level for the first time. That would show where spare capacity exists and where things are tight.

If new demand can be added without major upgrades, he argued, costs fall for everyone. In his words, a load-growth moment handled well could lower electricity costs rather than raise them.

The pattern across the interviews is that the technical ideas are known. What is missing is the data, the incentives and the rules to put them to work at scale.