WAYNESBORO, Ga. >> Towering over the Savannah River in Georgia, the first nuclear reactors built from scratch in the United States in more than 30 years illustrate the enormous promise of nuclear power — and its most glaring weakness.

The two new reactors at the Vogtle nuclear power plant will join two older units to create enough electricity to power 2 million homes, 24 hours a day, without emitting any of the carbon dioxide that is dangerously heating the planet.

But those colossal reactors cost $35 billion, more than double the original estimates, and arrived seven years behind schedule. That’s why no one else is planning to build large reactors in the United States.

Instead, the great hope for the future of nuclear power is to go small.

Nearly a dozen companies are developing reactors that are a fraction of the size of those at Vogtle, betting that they will be quicker and cheaper to build. As the United States looks to transition away from fossil fuels that have underpinned its economy for 150 years, nuclear power is getting renewed interest, billions of dollars from the Biden administration and support from Republicans.

One reason is that nuclear plants can run at all hours, in any season. To those looking to replace coal and gas with wind and solar energy, nuclear power can provide a vital backstop when the air is calm or the sky is cloudy.

But the push to expand nuclear power, which today supplies 18% of electricity, faces enormous hurdles.

In a major setback last week, the first serious effort to build small reactors in the United States was abruptly canceled amid soaring costs. While other projects are moving forward, the industry has consistently struggled to build plants on time and on budget. The Nuclear Regulatory Commission, which oversees the safety of the nation’s nuclear fleet, is less experienced with novel reactor technologies.

In 2008, utilities in Georgia and South Carolina sought approval to build two large reactors apiece, using a novel design with advanced safety features.

Construction began before designs were finalized and major changes had to be made partway through. Components arrived late. Workers installed 1,200 tons of rebar in a way that differed slightly from the design, triggering a 7 1/2 month regulatory delay. In 2017, South Carolina’s utilities abandoned their project after spending $9 billion with nothing to show except higher consumer bills. One utility went bankrupt and two executives pleaded guilty to fraud.

Only Georgia pushed ahead. Southern Company, the project’s largest owner, says the reactors will displace coal, which makes up one-fifth of its electricity mix.

To control costs, developers of next-generation reactors want to create smaller, standardized designs that require a lower upfront investment and can be easily duplicated.

“These nuclear megaprojects had just gotten way too complex,” said Jay Wileman, president of GE-Hitachi Nuclear Energy, which is designing a slimmed-down version of its boiling-water reactor that is only 300 megawatts — about one-quarter the size of the 1,117-megawatt units at Vogtle.

Ontario Power Generation plans to deploy four of them in Canada, hoping to bring down costs as it builds the same design again and again. The Tennessee Valley Authority is considering at least one.

Other companies are exploring radically new reactor designs that, in theory, can’t melt down and don’t require big containment domes or other expensive equipment.

Today, every American nuclear plant uses light water reactors, in which water is pumped into a reactor core and heated by atomic fission, producing steam to create electricity.

For nearly five decades, the Nuclear Regulatory Commission has regulated large light-water reactors. Now it has to consider a dizzying array of new technologies and their safety characteristics.

To date, the NRC has certified only one small reactor design, developed by NuScale Power. NuScale’s light-water technology is similar to existing plants, but the company argued that smaller reactors required different safety rules, such as smaller evacuation zones in case of accidents. Securing approval took a decade and cost $500 million.

The NRC says it is improving and recently greenlit a novel test reactor by Kairos, a startup, in just 18 months.

The ultimate test will be whether any new reactors get built.

Last week, NuScale announced it was canceling plans to deploy six 77-megawatt reactors in Idaho by 2030, which would have been the nation’s first small nuclear plant.

The problem was that it couldn’t sign up enough customers. Soaring costs didn’t help: In January, NuScale said the price of building the reactors had jumped from $5.3 billion to $9.3 billion, citing higher interest rates and materials costs.

Others say NuScale faced unique risks in going first, from navigating regulators to finding new business models.

“You see this a lot with emerging technologies,” said Joshua Freed, who leads the climate and energy program at Third Way, a center-left think tank. “Most of the early electric vehicle startups didn’t succeed, apart from Tesla. But electric cars are very much here today.”