TerraPower is a bold bet in an industry not known for entrepreneurial gumption. The nuclear-power start-up, based in Bellevue, Washington, is chasing an endless supply of affordable electricity for every person on the globe. That it would also greatly reduce supplies of depleted uranium and potentially eliminate a terrorist threat is fusionable gravy.

TerraPower is developing an innovative technology called a traveling wave reactor, or TWR, which runs on depleted uranium. The TWR requires only a tiny amount of enriched material to start up. An ensuing wave reaction burns depleted uranium slowly, over decades, with no need to refuel. (Conventional reactors require refueling every 18 months to two years.) Less-frequent refueling gives operators better control of their costs. And opening the reactor fewer times reduces opportunities for theft by terrorists or nuclear blackmailers.

Depleted uranium is one fuel source not in short supply: The Department of Energy estimates there are 686,000 metric tons in the United States alone. "There's a field in Paducah, Kentucky, that has 36,000 canisters of the stuff lying there," says TerraPower's CEO, John Gilleland. "There's enough to run the country for hundreds of years, I think."

Another improvement: The reactors use as a coolant liquid sodium, which is safer than water. And in an event like last year's earthquake and tsunami in Japan, the reactors would automatically shut down and continue to cool. "There's this wheel of desire," says Gilleland. "People say, 'I want energy security. I want safety. I want to reduce weapons.' We all start somewhere on that wheel."

Bill Gates and Nathan Myhrvold, Microsoft's former CTO, set the wheel turning in 2007. They were brainstorming in a lab at Intellectual Ventures, an intellectual-property incubator and licensor co-founded by Myhrvold. With them was renowned astrophysicist Lowell Wood. "We started out looking at energy in general and the huge deficit facing the world in the next century and beyond," Myhrvold recalls. "We knew we had to think about delivering massive amounts of emission-free energy with base-load power--meaning it's available every hour of every day year round, not something that goes out when the sun or moon passes the horizon or the wind dies down." Those conditions ruled out everything except nuclear, and the three began imagining designs that were technically feasible but not operationally practical. "We took those far-out ideas and dialed them back to something that was still a breakthrough and a remarkable leap forward but would also be licensable within a reasonable commercial life cycle," says Myhrvold.

To run the start-up, the three recruited Gilleland, founder of Archimedes Technology Group, a developer of solutions for disposing of nuclear waste. Sophisticated computer modeling proved the concept viable, and then things started happening fast, because TerraPower, unlike most nuclear ventures, required no public funding. The company now has around 60 employees. "The nucleus is getting interesting again," says Gilleland.

Gilleland says the first prototype will be ready for demonstration around 2022, a reasonable timeline given the technological and licensing hurdles. "It was interesting to watch Nathan and Bill when we said we can do this really fast--in eight to 10 years," says Gilleland. "In software, the scale for accomplishment is eight to 10 weeks." The company won't know if it has truly succeeded until the prototype runs for a few years, long enough to measure the resiliency of the materials and the efficiency of the fuel TerraPower is developing. Research and development is expected to cost as much as $5 billion.

Because TerraPower's approach is new, it must develop not only its own technology but also a supply chain for the components of that technology. The company needs sources for its distinctive fuel assemblies, which require nontraditional materials and manufacturing techniques. To develop that chain and conduct additional research, it is working with close to 100 partners, including national labs like Los Alamos ("We are actually feeding the U.S. government money," says Gilleland), schools such as M.I.T. and the University of Michigan, and companies such as Toshiba and Kobe Steel in Japan.

Licensing nuclear reactors is a high hurdle under normal circumstances; it is many times higher when the Nuclear Regulatory Commission, or NRC, lacks the expertise to evaluate what you are doing. TerraPower has talked informally with the NRC for two years but doesn't expect the organization will be ready to review and license the TWR until sometime in the 2020s. That creates a Catch-22: The NRC is unlikely to devote many resources to the TWR until TerraPower has a utility-company customer. But utility companies are unlikely to adopt the new technology unless they know they can get a license from the NRC.

So the company faces the further challenge of persuading another country to host its pilot, with the goal of producing data that can then help inform domestic regulators. TerraPower needed approval from five U.S. cabinet-level agencies before initiating high-level talks with China, India, Russia, South Korea, and a handful of other countries. "It's complicated and time-consuming but entirely appropriate," says TerraPower's senior technical adviser, Roger Reynolds. "It allows everyone to be certain we are not dealing in technology that is weapons related." (The company expects to sign an agreement with a foreign government next year.)

Gilleland says the State Department has encouraged export of the technology, because it will reduce the incentive for other countries to build their own enrichment plants. "We've had the support of the U.S. government to do this, because it makes the whole damn globe better off," says Gilleland.