Hyundai's 25,000-Atlas Plan Has No Yearly Rate

Boston Dynamics and Hyundai Motor Group opened a training facility for the Atlas humanoid robot at Hyundai's Metaplant America, outside Savannah, Georgia, the company said Sept. 21. Hyundai and Kia plan to deploy 25,000 Atlas units across their global plants over the next few years, according to The Robot Report, and a separate Hyundai-built U.S. factory is meant to manufacture up to 30,000 robots annually by 2028.

What Atlas is actually doing at the Metaplant Application Center

The Robotics Metaplant Application Center, or RMAC, is a testbed embedded inside Hyundai's own automotive plant rather than a Boston Dynamics lab. Atlas is now sequencing and staging automotive parts ahead of assembly, including lifting heavy loads and repetitive placement tasks, Boston Dynamics said. Component assembly is planned by 2030, a six-year gap between this week's opening and the point at which Atlas is trusted to actually build part of a car rather than hand one to a person who does. Zack Jackowski, Boston Dynamics' chief product and technology officer, called RMAC a "Robot Behavior Factory" and said the company's production-Atlas design assumptions have held up close to plan so far, a claim worth logging for when Atlas's task list gets harder.

RMAC is not staying this size. Boston Dynamics said the current facility will move into a new building about 10 times larger in 2027, and that the expanded site will start exploring Atlas applications outside automotive: aerospace, semiconductors, logistics, food and beverage, and life sciences. That sequencing matters more than it looks. A company confident in a design usually scales the same task across more sites before it scales the task list across more industries; going wider in 2027 while component assembly is still three years off suggests Boston Dynamics is treating the industry expansion and the harder manufacturing tasks as separable bets rather than one program that has to clear automotive first. Hyundai-affiliated entities own all of Boston Dynamics: Hyundai bought an 80% stake from SoftBank in 2021, and SoftBank sold its remaining 9.65% for roughly $325 million in a deal reported July 16, 2026. That full ownership is the reason RMAC sits inside a Hyundai plant rather than a neutral test site, and it means the 25,000-unit figure is, in part, a parent company buying its own subsidiary's output.

The number that matters is the missing one

The 25,000-unit figure is the one every outlet is running, but it is not the number that tells you what this deployment costs Hyundai or when it pays back. Twenty-five thousand robots over an unspecified number of years, deployed into an unspecified mix of plants, is a target, not a production rate. Compare that with how the rest of the humanoid field is talking about its own factories, as compiled by The Robot Report:

Company	Factory	Public annual target
1X Technologies	San Carlos, California	100,000+ NEO units by late 2027
Tesla	Fremont, California	50,000+ Optimus units per year
Agility Robotics	Salem, Oregon	10,000 Digit units per year at full capacity
Engine AI	Shenzhen, China	10,000 units per year
Hyundai / Boston Dynamics	Georgia, U.S.	30,000 units per year by 2028, no start-year figure given

Hyundai's 30,000-a-year 2028 figure is the closest thing to a run rate in its own announcement, and it is a capacity claim about a factory that does not exist yet, not a count of robots actually going into service. Every other company on that list names either a per-year rate already running or a specific date it expects to hit one. Hyundai has named a capacity and a deadline two years out; whether the 25,000 already committed to plants arrive before, during or after that factory comes online is not addressed in what either company has published.

What the design-parity claim is actually worth

Jackowski's comment that Atlas's production design has tracked close to the assumptions Boston Dynamics made is the kind of statement that sounds like nothing and is actually the whole story, because humanoid programs have a long history of exactly the opposite: a robot that works in a demo and then needs a hardware revision once it meets a real duty cycle. If that holds through the jump from logistics staging to the heavier assembly work planned for 2030, RMAC is a genuine manufacturing input. If it doesn't, the 25,000-unit figure is a procurement target that gets quietly revised the way safety-testing claims for other physical-A.I. systems have needed revision once they left controlled conditions. Boston Dynamics has not published failure or downtime data from the Metaplant deployment, and neither company said how many Atlas units are actually running at RMAC today, as opposed to the 25,000 planned over the coming years.

What would change this read is a per-year number for the 25,000-unit rollout, the same kind of figure autonomous systems in other domains are increasingly expected to report alongside a capability claim. Until Hyundai or Boston Dynamics gives one, the honest comparison to the rest of the field is a factory with a 2028 capacity target sitting next to four competitors that are already counting robots per year, and a customer that is also the factory's majority owner, which is not a comparison the rest of the field gets to make about itself.