Humanoid robots remain unproven at scale and have not yet achieved widespread commercialization, compared with industrial robots that have decades of proven track records in warehouses and factories. That frank acknowledgment is among the risks outlined in a securities filing by Agility Robotics—the first dedicated developer of humanoid robots to become a publicly traded company in the United States.

The robotics company, headquartered in Salem, Oregon, previously became the first to achieve full-time commercial deployment of humanoid robots at a customer warehouse in 2024. Now, Agility’s public listing on the Nasdaq stock exchange as of November 5 has come with financial disclosures detailing how the company plans to build a business around renting out or selling humanoid robots to perform manual labor.

“When we are doing robots as a service, it is much lower risk for customers to use it because they are getting a return on investment on day one,” Jonathan Hurst, co-founder and chief robot officer at Agility Robotics, told Ars. "Because they're paying a rate that is comparable to the human labor rate, and that makes the economics pretty straightforward.”

Agility is one of many startups attempting to commercialize humanoid robots. But the company has stood out from the competition through its lead in deploying and operating Digit humanoid robots with multiple customers, including the warehouses of logistics company GXO, the South Carolina factory of German automotive manufacturer Schaeffler, and the automotive production lines of Toyota Motor Manufacturing Canada. As Agility’s very first customer, GXO announced it would expand deployments of Digit robots beyond a warehouse in Georgia to additional facilities in North Carolina and Tennessee.

The filings accompanying Agility’s journey toward becoming a publicly traded company—based on a merger with a special purpose acquisition company—also provide a rare look at the business prospects and potential unit economics of humanoid robots. Most of its startup competitors remain privately held companies that have not been nearly as forthcoming.

Another exception has been China's humanoid robotics company Unitree, which opened its books when it went public on the Shanghai Stock Exchange STAR Market in August 2026.

Robot rental or ownership

Agility described the “illustrative unit economics” for both its “robots-as-a-service” subscription business and an “ownership model,” which would involve corporate clients purchasing robots outright, in an S-4 registration statement filed with the US Securities and Exchange Commission on September 4.

With the subscription model, customers would pay a one-time cost of about $25,000 per robot for delivery and deployment to a work site, along with a recurring monthly fee of $8,500 for software and maintenance.

Customers can also purchase the robot’s hardware outright for a one-time payment of $200,000, along with another one-time cost of $20,000 for delivery and deployment. But that still comes with an annual subscription fee of $36,000 for software and maintenance, or about $3,000 per month.

In exchange for renting or purchasing the robots, customers would get access to the latest Agility robot, known as Digit 5, starting in 2027. The bipedal robot has swappable end-effector “hands,” can lift up to 50-pound loads, and can perform more than 20 hours of work in a 24-hour period with the help of a fast-charging battery system.

The fast-charging battery technology provided by the UK-based company Nyobolt requires just nine minutes to deliver 90 minutes of robot runtime. “We worked with them for well over a year to design the pack that goes in our robot,” Hurst told Ars. “So the battery pack is fully custom and is ours, and we have the exclusive agreement with Nyobolt around using their [battery] chemistry in humanoid robots.”

The Digit 5 also comes with an onboard safety system that enables it to detect humans moving nearby and take appropriate actions, like standing still or squatting to minimize the risk of harm. That could unlock more opportunities to deploy Digit 5 beyond the confines of isolated work cells within warehouses and factories, said Pras Velagapudi, chief technology officer at Agility, in a previous conversation with Ars.

During a presentation for analysts and investors on October 6, Agility’s chief financial officer, Michael Beer, described the Digit subscription model as allowing customers to replace $200,000 in annual labor costs with $100,000 in annual robot costs, while buying a Digit robot outright could replace $1.1 million in labor costs with about $400,000 in robot costs over five years.

Agility’s business costs and revenue

Meanwhile, Agility described its own annual operating costs per robot as being $15,000 to provide the necessary software and maintenance, along with a one-time cost of $15,000 for Agility to initially deploy each robot.

The robotics company also estimated each Digit v5 humanoid robot as initially costing Agility $150,000 based on the bill of materials. Each robot is estimated to have a useful operational lifespan of about five years.

But Agility anticipates that bill of materials cost falling to $75,000 per robot once it ramps up to annual production of 1,000 robots, and eventually $30,000 per robot if the company can produce 10,000 robots per year. The company plans to gradually ramp up production at its RoboFab manufacturing facility in Oregon.

When balancing such business costs versus revenue, Agility anticipates that it could earn about $500,000 per robot on an annual basis through the subscription model, and about $400,000 per robot each year if a customer purchases the Digit robots outright while paying ongoing software and maintenance fees.

The subscription revenue calculations disclosed in Agility’s S-4 filing were based on an unidentified customer’s $300 million order covering 1,000 Digit 5 robots over a three-year subscription contract.

However, such a customer order does not mean Agility is guaranteed to earn $300 million, because the contract requires the robotics company to deliver on certain “contractual milestones, product features and specifications.”

The robotics company’s financial disclosures also bluntly state that “Agility has incurred net losses since inception and expects to continue to incur operating and net losses for the foreseeable future.” In 2025, the company earned just $1.8 million in revenue against $140 million in operating loss, which shows it’s still in the early stages of building a sustainable business.

Future risks and prospects

Agility said it continues to devote “substantially all of its resources to research, development, and commercialization of its humanoid robot,” while continuing to expand the RoboFab manufacturing facility. On October 21, Agility also announced plans to build a new facility in Pittsburgh, Pennsylvania, for supporting Digit robot deployments and further developing the robot’s software and AI.

Beyond the challenges of building its own business, Agility’s financial disclosures also cautioned that “competition in the humanoid and robotics market is intensifying rapidly, with well-funded entrants including Boston Dynamics, Tesla, Apptronik, Figure AI, 1X Technologies, Sanctuary AI, Hexagon, and NEURA Robotics, among others, actively developing competing humanoid systems.”

Many of those companies have advantages, such as “significantly greater brand recognition, customer relationships, and financial, technical, manufacturing, and other resources,” Agility’s S-4 filing acknowledged. It also listed business risks involving uncertain AI-driven robotic performance in the real world, safety issues in sharing workspaces with human workers, changing regulations, and cybersecurity threats.

Still, Agility co-founder Hurst described a large potential market with “millions of unfilled roles” involving manual tasks that would typically require higher-cost but relatively low-skilled human labor. Within the next year, Hurst expects multiple Agility customers to be sharing videos of the Digit robots “just very boringly doing their work” to the point where “all of the excitement and hype and novelty of humanoids has worn off.”

“It’ll look a bit like the dawn of the automotive market, and different companies will have different niches—some companies will build the Ford F-150 and some companies will build a Mazda Miata,” Hurst told Ars. “There's going to be a bunch of different niches for multi-purpose human-centric machines like these.”

Although Hurst sees Agility’s path eventually leading to a general-purpose autonomous robot in the future, he has consistently dismissed the idea of any company immediately deploying humanoid robots to riskier and less structured environments, like people’s homes. “Digit 5 is not going to walk into your living room and do dishes and stuff like that,” he said. “Nobody is.”

Whatever happens with Agility’s journey in the coming years, it will be instructive for anyone interested in the broader trajectory of the humanoid robot industry.

Read full article

Comments

During World War II, physicists on the Manhattan Project worried that the Nazi German nuclear program might achieve the first nuclear bomb before the US and its allies. That's not what transpired, but historians have debated just how close the Germans came to achieving criticality in their last attempt at building a reactor: the B8 pile. Werner Heisenberg, who led that effort, claimed that they were close and had only needed a little more uranium. But according to a paper published in the Proceedings of the National Academy of Sciences Nexus, that was not the case: the project simply didn't have sufficient uranium and heavy water to build a working bomb, given the design they were using.

This particular research started in 2013, when University of Maryland physicist Timothy Koeth—a co-author on the new paper—received the gift of a mysterious heavy metal cube from a friend. He instantly recognized it as one of the uranium cubes used by German scientists in their unsuccessful attempt to build a working nuclear reactor. Only 14 such cubes are known to exist today.

In 2019, we reported on Koeth’s quest to track down the cube’s origins, as well as several other similar cubes that had somehow found their way across the Atlantic. At the time, Koeth and his partner in the quest, then-graduate student Miriam “Mimi” Hiebert, had tracked down ten such cubes around the US. For instance, the Smithsonian Institute had a German uranium cube in storage. They tracked a third cube to Harvard University, where it was regularly passed around to students in introductory physics classes as a curiosity.

Underpinning the Manhattan Project in the US was the fear that German scientists under Adolf Hitler’s Nazi regime would beat the Allies to a nuclear bomb. The Germans had a two-year head start, but according to Koeth, “fierce competition over finite resources, bitter interpersonal rivalries, and ineffectual scientific management” resulted in significant delays in their progress toward achieving a sustained nuclear reaction. German nuclear scientists were separated into three isolated groups based in Berlin (B), Gottow (G), and Leipzig (L).

Heisenberg headed the Berlin group, and as the Allied forces advanced in the winter of 1944, Heisenberg moved his team to a cave under a castle in a small town called Haigerloch—now the site of the Atomkeller Museum. That’s where the group built the B8 reactor. It resembled an “ominous chandelier,” per Koeth, because it was composed of 664 uranium cubes strung together with aircraft cable and then submerged in a tank of heavy water shielded by graphite to prevent radiation exposure.

A replica of the failed B8 reactor on display at the Atomkeller Museum in Haigerloch, Germany Credit: ArtMechanic/CC BY-SA 3.0

As the German scientists were racing against time, Manhattan Project lead Lieutenant General Leslie Groves kicked off a covert mission dubbed “Alsos,” with the express purpose of gathering information and materials related to Germany’s scientific research. When Allied forces closed in, Heisenberg took apart the B8 experiment and buried the uranium cubes in a field, ferreting away key documentation in a latrine. (Pity Samuel Goudsmit, the poor physicist who had to dig those out.) Heisenberg escaped by bicycle, carrying a few cubes in a backpack.

A critical reassessment

As Heisenberg reported, the German scientists’ final experiment failed because the amount of uranium in the cubes was insufficient to trigger a sustained nuclear reaction. But Heisenberg was confident that “a slight increase in its size would have been sufficient to start off the process of energy production.” A model described in a 2009 paper supported that, showing the group would have needed only 50 percent more uranium cubes to make the design work.

Koeth and Hiebert uncovered a box of declassified documents about German uranium in the National Archives and found about 400 other uranium cubes from a separate reactor experiment by the Gottow group. They concluded that the combined inventory would have been more than enough to achieve criticality in the B8 reactor, suggesting that Germany’s secretive, isolationist approach hampered their nuclear program because the two groups weren’t sharing information or resources.

This latest paper builds on that earlier work but reaches a different conclusion. Koeth, Hiebert, and their co-authors analyzed two of the surviving uranium cubes, along with combing through historical records, technical reports, correspondence, and similar archival sources to learn more about the uranium, heavy water, and graphite blocks used for the B8 reactor experiment.

Among their findings: The B8 reactor would have needed between 1,292 and 1,775 uranium cubes to achieve criticality, and twice the heavy water produced by the Norsk Hydro plant in Vemork, Norway—the only source at the time—over the course of the entire war. In short, "Our calculations suggest Heisenberg significantly underestimated the materials required," the authors wrote.

The authors also cited prior work noting key differences in American and German graphite production. The former used petroleum coke. While some of this supply reached Germany, the British naval blockade in January 1940 deprived the Germans of that vital resource. The German nuclear physicists had to use metallurgical coke from the Ruhr, which had twice as many impurities as American coke. (The US scientists had similar issues until they located an unusually low-boron coke in Pennsylvania.) Thus, "no natural uranium, graphite-moderated pile could have gone critical" with the materials available to the German scientists at the time.

PNAS Nexus, 2026. DOI: 10.1093/pnasnexus/pgag282 (About DOIs).

Read full article

Comments

Caracas, Venezuela, earlier this year. Venezuela’s Caribbean coastline and land borders with Colombia, Brazil and Guyana make it an attractive location for potential U.S. troop deployments.

The more people use AI features in apps, the more the companies selling those apps have to spend on servers to run them, potentially squeezing their profit margins.

But some firms have found ways to keep costs in check. For instance, Microsoft-owned LinkedIn says it has flattened the cost of running its AI-powered Hiring Assistant, which uses large language models to automate candidate search and outreach for recruiters, even as usage of the tool has steadily increased in the past year. 

Studies have shown that the betting sites can predict races correctly, but their ability to forecast U.S. midterm elections is an important proving ground.

A war room. A global conference call. A founder who live-posted about the largest crypto heist in history. Inside a frantic 24 hours at the world’s biggest crypto exchange.