Global Humanoid Robot Shipments Nearly Triple in H1 2026, Led by China’s Agibot
Agibot shipped somewhere between 8,400 and 9,700 humanoid robots in the first half of 2026, enough to control more than 40 percent of a market that barely existed at scale eighteen months ago. That number is real. The claim now traveling alongside it, that Hyundai and other major automakers are already putting humanoid robots to work on their assembly lines, is not.
The Shipment Numbers, and Why Trackers Disagree
Counterpoint Research puts Agibot’s first-half shipments at roughly 9,700 units, more than 43 percent of a global market that topped 22,000 units, up nearly 300 percent year over year. Smart Analytics Global (SAG) counts it differently: about 8,400 units and a 44 percent share, out of roughly 19,100 units shipped industrywide, up 272 percent from the 5,100 units the same firm counted a year earlier. Unitree Robotics, the other Chinese manufacturer most investors watch, shipped roughly 5,900 units for a 31 percent share, according to SAG’s count.
The gap between trackers isn’t a scandal. It reflects a market still deciding what counts as “shipped,” whether that means units delivered to a paying commercial customer or units sent to a research lab or demo program. What both trackers agree on matters more than the few thousand units of daylight between them: Chinese vendors now account for more than 97 percent of global humanoid robot shipments, and Agibot has overtaken Unitree, long seen as the category’s early leader, for the top spot. Unitree priced its IPO in August, the same week SAG’s count confirmed it had slipped into second place.
Nvidia’s “ChatGPT Moment” Line, in Context
Nvidia CEO Jensen Huang gave the industry its rallying line at CES 2026 in Las Vegas on January 5. “The ChatGPT moment for physical AI is here,” he said, “when machines begin to understand, reason and act in the real world.” He said it while unveiling Alpamayo, Nvidia’s new AI model for autonomous vehicles, and named robotaxis, not humanoid robots, as the technology’s first real beneficiary. Nvidia sells the chips that power both categories, which doesn’t make the line wrong. It does mean the company with the most to gain from a physical AI narrative wrote its headline, and the headline was made for a different product than the one now being quoted to justify it.
Hyundai’s 25,000 Robots Don’t Exist Yet
Hyundai Motor Group has committed to deploying more than 25,000 Atlas humanoid robots, built by its Boston Dynamics subsidiary, across its manufacturing operations. That’s a real number, but it’s a 2028 number. The company plans to start at its Metaplant America factory in Georgia that year, expand to Kia’s Georgia plant in 2029, and consider external sales only after its own factories are running. Hyundai announced the plan at a JPMorgan investor conference in Boston in May, not from a production floor. No Atlas units are deployed in any Hyundai factory today.
The economics explain the delay. Atlas currently costs an estimated $130,000 to $140,000 per unit to build. Hyundai expects that to fall to around $30,000 once cumulative production passes 50,000 units, the kind of manufacturing scale that doesn’t happen by simply announcing an intent to reach it. Hyundai Mobis, the group’s parts division, is building a US actuator plant meant to reach 350,000 units of annual capacity starting in 2028. That’s an infrastructure bet, not a deployment.
What’s Actually Running on a Factory Floor Right Now
Some humanoid robots are working in Western factories today, just not at Hyundai’s scale or under its name. BMW’s plant in Leipzig, Germany began testing AEON, a humanoid built by Zurich-based Hexagon Robotics, in December 2025. AEON delivers materials to the assembly line and handles high-voltage battery component work, and BMW plans to move it into full pilot operations this summer. It follows an earlier BMW pilot at the company’s Spartanburg, South Carolina plant, where Figure AI’s Figure 02 robot assisted with more than 30,000 vehicle builds in 2025.
Those are pilots, not fleets. Neither program comes close to the volume Hyundai has promised for 2028, and BMW hasn’t announced plans to scale either robot into a full production role. But they’re the closest thing to Hyundai’s stated vision that currently exists outside China, running today instead of sitting on a roadmap.
The Reliability Gap Nobody’s Solved
Industry veterans are less convinced the timeline is the real obstacle, because the technology isn’t there yet. Jim Brown of Teradyne Robotics, which makes industrial automation systems, put it plainly: “There’s definitely a lot of buzz,” but robot makers aren’t always “thinking about what problems we are actually trying to solve.” He points to a cost most demos skip past: standing still isn’t free. Today’s humanoid robots draw meaningful power just holding a pose, a real cost on a factory floor that already runs wheeled robots and fixed arms that solve the same tasks more efficiently.
General Motors’ Mikell Taylor frames the reliability bar more bluntly: “80% success is failure in production.” A robot that works four times out of five in a lab demo doesn’t survive contact with a line that has to run the same way every shift, without an engineer standing by to intervene. That gap, not the shipment count, is what decides whether 2028 turns out to be an early estimate or an optimistic one.
China’s manufacturers have already answered whether humanoid robots can be built and sold at scale. What they haven’t answered, and what Hyundai’s 2028 target and BMW’s slow pilot expansion suggest nobody has, is whether those robots can run a Western factory floor reliably enough to replace the wheeled robots and fixed arms already doing the job. The shipment numbers point to a manufacturing base that’s ready. The deployment numbers point to buyers who still aren’t sure.
The post The Humanoid Robot Boom Is Real, The Factory Takeover Story Isn’t, Yet appeared first on DataFLOQ.
