Thirty Thousand Robots a Year From 2028

Hyundai Motor completed full ownership of Boston Dynamics in July 2026 by buying SoftBank's remaining shares for approximately $325 million, on top of the 80 percent stake acquired in 2021 for $1.1 billion. It opened a Robot Metaplant Application Center in Savannah, Georgia in June 2026 and plans to expand it tenfold by the end of the year, while searching for a US site for a dedicated robotics manufacturing facility. US production is targeted for 2028 at 30,000 units a year, with guaranteed customer orders already covering 25,000 units. Atlas humanoids are scheduled to begin parts sequencing at Metaplant America in 2028, advancing to component assembly by 2030. Boston Dynamics' product line comprises Atlas, Spot and Stretch. Chief executive José Antonio Muñoz is quoted; automotive revenue is close to 80 percent of Hyundai's total.

The two dates are what make this worth reading rather than another humanoid announcement. Parts sequencing in 2028 and component assembly in 2030 is a deliberately conservative sequence, and the choice of first task tells you what is actually achievable. Sequencing — collecting the right parts in the right order for a station — is a pick-and-place problem with known parts, a structured environment and a tolerance for taking longer than a person would. Component assembly requires force control, tolerance to variation and recovery from a failed insertion, which is a substantially harder problem. Setting them two years apart is an honest engineering judgement rather than a marketing one.

The 25,000 guaranteed orders against 30,000 units of capacity is the number that separates this from the rest of the humanoid sector, and it is worth being careful about what it means. Hyundai is both manufacturer and customer: the largest guaranteed buyer of these robots is a car company that owns them, deploying into plants it controls. That is a genuine advantage — it removes the chicken-and-egg problem of building a factory for a product nobody has bought — and it is also not the same as external demand. The test of the programme is whether anyone outside Hyundai orders at volume, and that will not be visible until after production starts.

For manufacturers watching this sector the useful question is not whether humanoids work but where they would be cheaper than the alternative. A humanoid's argument is that it fits an environment built for people without modifying it, which matters in brownfield plants with narrow aisles, mixed tasks and frequent changeover. Where a task is repetitive, high-volume and fixed, a conventional robot arm bolted to the floor will beat it on cost, speed and reliability for the foreseeable future. The interesting deployments will be the ones in between — and the fact that Hyundai's own first application is sequencing rather than assembly suggests they see the boundary in the same place.