The Iron
Workers

The robot had been working at the BMW assembly plant in Spartanburg, South Carolina for six months when the engineers stopped noticing it. That's the real inflection point: not when a machine does something spectacular on stage, but when it becomes unremarkable on a factory floor.

In 2026, that moment arrived. Humanoid robots are no longer a demo-room phenomenon. They are deploying — expensively, gradually, with plenty of caveats still attached — but deploying in real industrial environments, performing real productive tasks, and generating actual revenue for the companies that build them. The transition from pilot to platform, which the industry has been promising for years, is finally, tentatively, beginning.

The Factory Floor

Figure AI's Figure 02 robot has been operational at BMW's Spartanburg, South Carolina plant since mid-2025, performing tasks that include inserting sheet metal components, sorting parts, and transporting materials between stations. Over its deployment period, Figure-equipped workstations contributed to the production of more than 30,000 BMW X3 vehicles. That's not a demonstration. That's manufacturing at commercial scale.

BMW is now expanding the partnership. Figure's next-generation Figure 03, running on Helix 02 — the company's full-body AI model — is being deployed at BMW's Leipzig plant, with 24/7 autonomous operation targeted in defined workstation scenarios. Figure AI scores 78.9 out of 100 in third-party commercial readiness assessments, well ahead of competitors. Its focus has been ruthlessly practical: pick a limited set of tasks it can do reliably, do them reliably, and expand the task set from a foundation of trust.

Boston Dynamics, meanwhile, has committed its electric Atlas humanoid to Hyundai Motor Group manufacturing facilities, with Hyundai investing in a dedicated robotics factory capable of producing 30,000 units per year. If that capacity is reached, it represents a genuine industrial workforce — not a pilot. Boston Dynamics brings two decades of bipedal locomotion research to Atlas, and the electric version demonstrated in 2024 showed a level of dynamic, self-correcting movement that earlier hydraulic platforms couldn't match. The question was never whether Atlas could move. The question was whether it could insert a bolt repeatably for eight hours.

The Chinese Surge

The American narrative of humanoid robots has a competitor that's winning on volume. China's approach has been characteristically different: faster, more numerous, and priced for a market that doesn't yet exist at full scale but that Chinese manufacturers are determined to create.

AgiBot, a Shanghai-based startup, produced its 10,000th humanoid unit in late March 2026 — scaling from 1,000 units in all of 2025 to 10,000 within months of the new year. Its revenue reached approximately $140 million in 2025 alone. A $200 million Series A round, one of the largest for any humanoid robotics startup, funded the expansion. AgiBot's market share of approximately 31% makes it the global leader by units shipped.

Unitree Robotics, the second-largest Chinese manufacturer, sold 5,500 humanoid units in 2025 and launched its G1 platform at a starting price of approximately $16,000. The shock of that number is hard to overstate. Current-generation Western humanoids range from $100,000 to $300,000 per unit. At $16,000, Unitree's robot costs less than a used car. It cannot yet perform the tasks that Figure 03 can perform at BMW. But it performs tasks that justify $16,000 in a very wide range of industrial and service environments — and the capability gap is narrowing.

Chinese companies controlled approximately 90% of the global humanoid robot market in 2025. That dominance reflects a policy environment, a manufacturing ecosystem, and a cost structure that Western competitors can't replicate in the near term. The US-China Economic and Security Review Commission documented this trajectory in a March 2026 report, describing China's "two loops" model: competitive domestic production builds scale, while export pricing disrupts global markets. The pattern is familiar; the product is new.

Why Japan Airlines Matters

In some ways, the most instructive 2026 deployment isn't a factory — it's an airport. In May, Japan Airlines deployed humanoid robots at Tokyo's Haneda Airport: two Unitree-based platforms, operated through a partnership with GMO AI & Robotics, priced at approximately $15,400 per unit under a three-year operational commitment.

Japan Airlines is not a startup performing a showcase. It's a century-old aviation carrier in one of the world's most safety-conscious regulatory environments, making a multiyear operational commitment. That means insurance, liability, regulatory approval, maintenance protocols, and integration with airport operations systems. None of these are trivial. JAL's decision reflects a calculated bet that the labor costs and workforce constraints in Japan's service economy make even today's limited humanoid robots economically defensible.

The tasks being performed at Haneda are not the stuff of science-fiction robot cinema. They are repetitive, low-complexity service tasks — the kind that nobody wants to do, and that staffing is hard to sustain in Japan's aging economy. That gap between "spectacular" and "useful" is the real story. The robots don't need to be brilliant. They need to be reliable, affordable, and present.

What They Can and Cannot Do

Honesty about current capabilities is essential to understanding the moment. The humanoid robots being deployed in 2026 can perform a well-defined set of tasks in structured, predictable environments. They excel at repetitive precision work with consistent objects — inserting panels, sorting items, transporting loads along fixed paths. They struggle with unexpected variations, cluttered or dynamically changing spaces, and anything requiring the delicate, adaptive manipulation that human hands perform without thinking.

Nvidia's Isaac GR00T open model platform, released in 2025, addresses the AI side of this constraint — providing a foundation model that enables robots to parse natural language instructions and sequence complex multi-step tasks without custom programming. But the operational infrastructure gap remains real. Persistent environment mapping, reliable network connectivity, charging logistics, remote monitoring, and integration with existing facility management software must all be built before a robot can be deployed. This is not a minor engineering detail; it is, in many cases, the main constraint on deployment speed.

The Economic Equation

The case for humanoid robots depends entirely on what happens to their price and capability curves over the next five years. At $100,000–$300,000 per unit, with narrow task profiles and significant infrastructure overhead, humanoids are a specialized industrial tool — valuable in specific niches but not economically transformative. At $15,000–$30,000 per unit with expanding task repertoires, the calculation changes fundamentally. Almost every labor-intensive industry becomes a potential market.

The price compression is happening, though not as fast as optimists predicted. AgiBot and Unitree are producing at scale in China with a cost structure that Western manufacturers cannot match. Tesla, with its manufacturing expertise, could theoretically compete on price if Optimus reaches production maturity. The more likely outcome over the next three years is a bifurcated market: Chinese manufacturers dominate volume and price-sensitive applications; American and European companies compete on capability, safety certification, and enterprise integration for high-value tasks in regulated industries.

What's clear is that 2026 is the inflection point — the year the category moved from impressive demonstrations to early production. The global humanoid robot market is projected to reach $4–5 billion this year, growing at a 39% compound annual rate through 2030. The robots are on the floor. They're doing the work. And we've almost stopped noticing. In the history of technological change, that's what arrival looks like.