Robotics —

Hyundai Tells JPMorgan, May 19: 25,000 Boston-Dynamics Atlas Humanoids Across Hyundai + Kia Plants, 30,000-Unit Annual Production Capacity by 2028, 300,000 Actuator Units Per Year Onshored to U.S. Facilities — Georgia Metaplant Lights Up in 2028, Kia Georgia 2029, Korean Union 「Demanding Labor Agreements First」

At a JPMorgan IR session on May 19, Hyundai put a number on its humanoid program: 25,000 Atlas robots across Hyundai/Kia plants, 30K-unit annual production capacity by 2028, 300K US-made actuators per year. The Korean union has already demanded a labor agreement.

By LostJobs Editorial
Hyundai Tells JPMorgan, May 19: 25,000 Boston-Dynamics Atlas Humanoids Across Hyundai + Kia Plants, 30,000-Unit Annual Production Capacity by 2028, 300,000 Actuator Units Per Year Onshored to U.S. Facilities — Georgia Metaplant Lights Up in 2028, Kia Georgia 2029, Korean Union 「Demanding Labor Agreements First」

The single most important sentence in a humanoid-robotics announcement is the one that names a unit count. On Tuesday, May 19, at a JPMorgan Chase investor-relations session, Hyundai Motor Group named one: 25,000+ Atlas humanoid robots to be deployed across Hyundai Motor and Kia manufacturing facilities, with annual production capacity of 30,000 units by 2028, and 300,000 actuator units per year to be onshored at U.S. facilities. The robots come from Boston Dynamics, which Hyundai bought from SoftBank in 2021 and which now sits inside Hyundai Motor Group as the robotics subsidiary.

The deployment schedule is the part the May 6 handstand demo did not specify. Per Kia CEO Song Ho-sung’s overseas roadshow material reflected in the JPMorgan deck, the first plant to receive Atlas at scale is the Hyundai Motor Group Metaplant America in Bryan County, Georgia — the multi-billion-dollar EV facility that started SUV production in 2024 — beginning in 2028. Kia’s Georgia plant in West Point follows in 2029. The other Hyundai/Kia U.S. and Korean plants are not given dates in this round of disclosure.

What “25,000 Atlas” actually buys at the cohort scale

For comparison: in May 2026, the next most-aggressive Western humanoid program is Figure AI’s 40-unit Figure 03 fleet at BMW’s Spartanburg plant, billed to BMW at roughly $25 per robot-operating-hour per LostJobs’ January coverage. Unitree shipped about 5,500 humanoid units across all customers in 2025 and is publicly targeting 20,000 units in 2026. Tesla’s Optimus V3 is now slated for late-July / August reveal with production to follow.

A 25,000-unit single-customer deployment plan inside a single automotive group, with the manufacturer also building 30K-per-year of its own assembly capacity, is a different order of magnitude. It is roughly equivalent to all of Unitree’s 2025 global humanoid output deployed inside one OEM’s plants — except the OEM in question owns the robot maker and is also onshoring the actuator supply. The 300,000-actuator-per-year number is the load-bearing detail: an Atlas-class humanoid uses on the order of 25–35 high-torque actuators (hips, knees, ankles, shoulders, elbows, wrists, neck, plus auxiliaries), so 300K/year of actuators is exactly the supply-chain throughput to support 8,000–12,000 Atlas units in steady-state production. That bracket is consistent with the 30K-annual-capacity claim once warranty replacement and OEM-export volumes are netted in.

In other words: this is not a press-release number padded for an investor day. It is the actuator math.

The 2028–2029 calendar, and the gap it opens

The detail the May 19 disclosure does not close is what happens in 2026 and 2027. Atlas’s production-spec May 6 handstand demonstration showed a unit that can do gymnastics in a controlled studio; the May 19 deck shows a unit that will start touching a Hyundai assembly line about 30 months later. Between now and 2028, every other entry in the cohort — Figure, Tesla, Unitree, Agility Digit at GXO Logistics, AGIBOT, Apptronik at Mercedes — is going to be shipping into customer pilots, and several of them will be doing it without Hyundai’s 30-month industrialization runway.

The strategic read this opens is the one auto-industry analysts have been quietly debating since the Bloomberg piece on the May 6 demo: Hyundai’s pace is slower than Figure’s or Unitree’s because Hyundai’s pace is conditioned by the full automotive supplier-qualification cycle. Atlas in 2028 will not be a pilot. It will be plant-qualified, line-rated, change-controlled, FMEA-documented, and OSHA-permitted. That takes 30 months because that is how long automotive line-readiness takes for any non-trivial mechanical system, robot or otherwise. The other humanoid programs are buying speed by accepting a slower path to that qualification posture.

By 2029, when Kia Georgia lights up, the question becomes which posture the labor-market data prefers: 40 Figure 03 units at BMW under a per-hour rental, or 25,000 plant-qualified Atlas units at Hyundai/Kia inside the supplier-qualification regime. The first is a service. The second is a substitute for headcount.

The labor side, already in motion

On the labor side the picture is as clear as the unit count. Per Korean union reporting tracked by US press, Hyundai’s South Korean unions have already demanded a labor agreement be negotiated before any large-scale humanoid deployment. The historical reference point is the 2017 Hyundai-union framework that capped the deployment of certain automation technologies pending bilateral agreement; the union is asking the same posture for Atlas, with the unit count now public.

Company-side messaging tracks the standard humanoid script. The official line is that the system is tightly controlled and intended to take on physically demanding or hazardous work rather than replacing human workers. The published Tier-1-CEO arithmetic that Wards Auto and the WSJ have both cited is less polite: one humanoid pays for itself in about two years even at a roughly $250,000 unit cost, because a robot can work two shifts and replace two workers. At 25,000 units, the arithmetic the labor side will run — irrespective of company assurances — is approximately a 50,000-shift-position substitute over the deployment horizon. Whether that number lands as net headcount reduction or as redeployment depends on the labor agreement the Korean and U.S. unions can extract before the 2028 first-light.

The unit count is now public, which is what makes the agreement possible to negotiate.

What to watch

  • The Hyundai-union framework. A Korean-union published agreement before end-Q4 2026 would set the deployment posture not just for Hyundai but for every other automaker watching this play out. No agreement, and the May 19 deck is the negotiating position.
  • The Georgia HMGMA plant 2027 qualification milestones. Atlas plant-qualification will happen at HMGMA in 2027 for a 2028 light-up. Any leak of which stations are being qualified (paint, body, assembly, materials handling) is the practical roadmap.
  • The actuator supplier list. 300,000 actuator units per year requires either Hyundai Mobis tooling up its own actuator line or qualifying a U.S. tier-1 (Allied Motion, Moog, Schaeffler North America). The supplier disclosure in the next Hyundai earnings call is the load-bearing detail for U.S. industrial-base implications.
  • Whether the 25K number ever appears in a Boston Dynamics public statement. As of May 19, Boston Dynamics has not separately confirmed the unit count in its own channels. A Boston Dynamics-side announcement would be the cleanest tell that the deployment plan is not just an IR aspiration.

The May 6 demo was a YouTube short. The May 19 deck is a number. That is the move the humanoid cohort has been waiting for one of the OEMs to make.

Sources