Atlas shows how deep tech survives

This year’s global boom in artificial intelligence robots has a clear star: Atlas, a humanoid robot. Boston Dynamics, the U.S. robotics company behind Atlas, has never posted a sustained profit since its founding in 1992. Until recently, it repeatedly fell into capital impairment. Even after Hyundai Motor Group acquired the company for about $1.1 billion in 2021, cumulative losses over the past four years have exceeded 1.2 trillion won.

Had the company been based in South Korea, it likely would have been labeled a zombie firm and forced out of the market within five years of its founding. It is difficult to imagine capital willing to wait 34 years for technology to fully mature. The breathtaking joint movements unveiled on the CES 2026 stage last month were the result of decades of investment and patience. How was such seemingly irrational deep tech survival possible in the United States?

Boston Dynamics began as a corporate spinoff in 1992, when Marc Raibert, then a professor at the Massachusetts Institute of Technology, incorporated his university research lab known as the Leg Lab. At a time when robots largely belonged to the realm of science fiction, the company’s foundation rested on government funding.

This support went far beyond one time startup subsidies. The Defense Advanced Research Projects Agency, or DARPA, under the U.S. Department of Defense, became the company’s first customer by issuing demanding research assignments. These orders were driven by forward looking imagination, including whether transport robots could be deployed to hazardous terrain in place of soldiers.

DARPA is widely regarded as the cradle of U.S. deep tech, channeling massive budgets into projects with a high likelihood of failure. Both the internet and voice assistant technologies trace their origins to the agency. In a 2016 white paper, DARPA officials wrote that if none of their projects had failed, it would indicate they had not taken enough risk. The agency has long tolerated failure rates of up to 90 percent.

With DARPA’s support, Boston Dynamics could pursue long term technology accumulation over a 20 year horizon, free from the pressures of short term profitability. BigDog, the quadruped robot unveiled in 2005, and the first bipedal Atlas model introduced in 2013 both emerged from extended research programs. U.S. Congress did not reprimand the agency for the absence of immediate commercial results.

By the 2010s, robotics technology had begun to appear on the investment radar of major tech companies. Although the market itself remained nascent, large corporations increasingly viewed robotics as a future growth engine. This shift was evident in Google’s $500 million acquisition of Boston Dynamics in 2013, signaling the creation of a “capital greenhouse.”

After being sold to SoftBank’s Vision Fund in 2017, the company ultimately found a home at Hyundai Motor Group in 2021, where manufacturing synergies became possible. The 34-year journey combined the efforts of universities, led by MIT, which planted the seeds; DARPA, which provided nurturing support; and global capital, which served as the greenhouse that transformed imagined technology into reality. Large scale production is now within reach.

South Korea presents a sharp contrast. University lab spinoffs are difficult to establish, and the government struggles to fund long term research projects without visible results. Acquisitions by large corporations are extremely rare. Only about 3 to 4 percent of South Korean startups achieve exits through mergers and acquisitions, leaving initial public offerings as virtually the only viable option. In the United States, by comparison, nearly 90 percent of startups are acquired.

Regulatory hurdles are a key reason large scale acquisitions remain challenging. When conglomerates acquire domestic startups, they face extensive regulations tied to affiliate designation. If performance falls short, executives risk accusations of breaching fiduciary duty, and the resale market is limited. With the nurturing infrastructure itself underdeveloped, launching a deep tech startup becomes nearly unthinkable.

The Lee Jae-myung administration recently held a startup audition program and declared its ambition to transform South Korea into an entrepreneurial society. While this is a positive step toward fostering a startup boom, it should not be overlooked that the growth of world class deep tech depends less on flashy events and more on patient capital willing to tolerate failure, supported by institutional infrastructure that makes such patience possible.