Hyundai commits to a 30,000-robot-a-year US plant — with 25,000 units already spoken for

Hyundai Motor used its 2026 CEO Investor Day, held at the Conrad Seoul in Yeouido on August 26, to put dates and volumes on a physical-AI plan it has been describing in general terms for a year. Chief executive José Muñoz said the group is “transforming into a physical AI company” and is targeting mass production and deployment of robotics ahead of competitors.

The specifics: Boston Dynamics’ Atlas humanoid goes onto production lines at Hyundai Motor Group Metaplant America (HMGMA) in Georgia at scale from 2028, expanding to global sites from 2030. In parallel, Hyundai will build a commercial robot manufacturing plant in the United States, in production from 2028 with annual capacity of 30,000 units. The site will be announced before the end of this year. Muñoz called it what he expects to be the world’s first large-scale commercial robot manufacturing facility, and said 25,000 units of initial demand are already confirmed.

The 25,000 number, read carefully

That confirmed volume starts with HMGMA and expands to external industrial and logistics customers afterwards. In other words, the anchor order is internal. Hyundai is both manufacturer and first customer, which is exactly how it de-risks a factory that would otherwise be built against a forecast.

This is not a criticism so much as the whole point. The humanoid sector’s recurring problem is that capacity announcements float free of any buyer — a monthly output target with no order book behind it. A vertically integrated group with its own assembly lines can commit to consuming its own output, and 25,000 units is enough volume to walk down a cost curve before a third party is asked to sign anything. What it does not prove is external demand at a price that clears. That test starts after HMGMA.

RMAC is the part that is already running

The bridge to 2028 is the Robot Metaplant Application Center, which Hyundai opened in the US in June 2026 to test manufacturing AI robots in an environment built to resemble a live vehicle plant. It is currently working across five vehicle models and 67 task items, covering sequencing, overhead work, high-repetition tasks and quality inspection. Hyundai will expand the RMAC site roughly tenfold by the end of this year, with the stated aim of accumulating millions of hours of real manufacturing data.

That sequencing — a data-gathering facility now, a deployment date in 2028, a plant date in 2028 — is more disciplined than most humanoid roadmaps, which tend to jump from demonstration to volume with nothing in between. The 67 task items are also the honest part of the announcement: the tasks are enumerated because they are the tasks the robots can currently be evaluated on, not the whole factory.

The robotaxi half arrives first

The autonomy side is nearer term. Hyundai will deliver Ioniq 5-based robotaxis built at HMGMA to Waymo in the second half of 2026; Muñoz said nine months of public-road testing are complete and the vehicles are ready for Waymo’s use this half. He also said Hyundai plans to extend the Waymo partnership into global markets. Separately, Motional — the group’s autonomy joint venture — intends to launch a driverless commercial ride service with Uber in Las Vegas before the year ends, hailed through the Uber app.

The framing Hyundai puts on this is an “AV foundry”: developing and building purpose-fit vehicles for autonomy companies and mobility platforms rather than operating the fleet itself. That is a contract-manufacturing position, with contract-manufacturing margins, but it is also the only position in the robotaxi stack where revenue does not depend on solving autonomy.

What Hyundai has that most robot companies do not

Muñoz was explicit that the group intends to own the full chain — development, production, distribution, sales and financing. The distribution network already exists through dealer partnerships, and Hyundai Capital is assessing financing products for robot sales.

That last item is the one worth watching. Industrial robots are sold on payback periods, and the buyers who most need them are the ones least able to fund them out of cash. Whoever attaches captive financing to humanoids first will find that the constraint on adoption was never entirely technical. Hyundai already runs a financing arm sized for vehicles, which is a materially different starting position from a robotics startup pricing per unit.

The caveats are the usual ones for investor-day commitments: 2028 is two years out, no plant site has been named, and Boston Dynamics has never manufactured Atlas at volume — Spot and Stretch, yes, a bipedal humanoid, not yet. Hyundai’s own robot programme has moved slower than announced before. But the number of companies that can credibly pair a humanoid, a factory to put it in, a plant to build it and a lender to finance it is very small, and this week it is one shorter than the number of companies talking about doing so.