May 2026 Humanoid Report: The OEMs Took the Story Back

The most important fact May produced for humanoid robotics wasn’t another Atlas handstand. It was Hyundai walking JPMorgan through a five-digit unit count: 25,000 Atlas robots, 30,000-unit annual production capacity by 2028, 300,000 actuators a year onshored to the U.S.

That one number changed the subject of the 2026 humanoid race from “can the robot makers build it” to “how do the automakers plan to schedule it.” Inside the same month: Tesla took the 14-year Model S line in Fremont apart to make room for Optimus, 1X opened the first vertically integrated humanoid factory on U.S. soil in Hayward, Figure’s BotQ shipped 240 units in April after doubling monthly for three months running, and Agibot’s G2 posted 99.9% task success and 310 units per hour on a Longcheer tablet line. The demo era is ending. The operational-metrics era starts here.

On the other side: Korea’s KAPEX project wrote down the number every other humanoid program has refused to publish — current state-of-the-art task completion is 30%, with three to four minutes of usable runtime per day. And on May 3 Xinhua front-paged a portrait of Agibot co-founder Peng Zhihui timed to the May Fourth Medal — the first humanoid founder ever logged into China’s highest youth state honor.

This report breaks both halves open: eight named signals from the month, three structural reads, two counter-narratives that aren’t going away. It closes with five leading indicators for June. Two of those will tell you whether OEM annexation is now the template, or just Hyundai running ahead alone.