Humanoid robots will go from a few thousand units a year to a mass-market product within fifteen years, according to a new forecast from Berg Insight. The Swedish research firm expects annual shipments to climb from 16,000 in 2026 to 26 million by 2040, a compound growth rate of 63.7 percent, with the market’s value rising from $890 million to $554 billion over the same period.
“The humanoid robotics industry has reached an important commercial inflection point,” said Berg Insight analyst Julia Wahlen. For anyone who builds or sells connected devices, though, the more telling number sits a little further down the report.
Why A Robot Needs A Mobile Plan
Berg Insight expects cellular subscriptions tied to humanoid robots to grow from about 5,000 in 2025 to 50.6 million by 2040, or 84.9 percent a year. The total outruns annual shipments because subscriptions build up as robots stay in service, so humanoids start to look like a long-lived connected fleet.
The reasons are familiar to anyone in IoT. The firm points to remote monitoring, fault diagnosis, fleet management, software updates and cloud services, all of which need a link that works wherever the robot happens to be. A humanoid walking between a warehouse floor, a loading bay and a car park can’t count on one Wi-Fi network the way a fixed industrial arm can.
Software updates matter most. A humanoid’s abilities depend on AI models that its maker will keep improving, so the robot sold in 2030 is expected to learn new tasks over the air for years afterwards. We looked at that shift from sensing to acting in how physical AI changes IoT, and connectivity firms have noticed too: Aeris recently hired a marketing chief to sell connectivity for physical AI.
The Starting Line Is Already Moving
Those connections start from a small base, but the base is growing faster than Berg’s own starting figure suggests. Research firm Omdia counted about 13,000 humanoids shipped worldwide in 2025, more than five times the 2024 total. Chinese makers led by a wide margin: AgiBot shipped over 5,100 units, Unitree about 4,200 and UBTECH around 1,000, while Figure AI, Agility Robotics and Tesla each shipped somewhere between 150 and 500.
This year the pace has picked up again. Smart Analytics Global put first-half 2026 shipments at roughly 19,100 humanoid units, with Chinese companies taking more than 97 percent, which is already above the 16,000 Berg Insight forecasts for the whole of 2026.
Research firms count differently, so the gap may partly come down to what each one treats as a humanoid, whether small research and education robots are included, and when a unit counts as shipped. Still, the direction is clear. The early part of the curve is steeper than the forecast’s baseline, and it is being drawn almost entirely in China.
Forecasts Agree On Direction, Not Size
Where the curve ends is far less certain, and the big banks show how wide the range is. Goldman Sachs has raised its estimate fivefold to 6.5 million humanoids in total by 2035, worth about $138 billion that year, citing faster AI progress and falling hardware costs.
Morgan Stanley looks further out and bigger, expecting close to a billion humanoids in use by 2050 and a $5 trillion market including supply chains and services. It also expects adoption to stay slow until the mid-2030s before speeding up, which is roughly the shape Berg Insight’s numbers follow.
Our read: every forecast points the same way, but they differ so much that 26 million a year is one plausible scenario, not a target. The near-term shipment numbers will show whether the curve bends on schedule.
What Has To Go Right
Bending that curve depends on more than demand. “Competition is increasingly shifting from hardware performance alone towards software, embodied AI and scalable manufacturing,” said Berg Insight analyst Emilia Rickard, and embodied AI here simply means AI that controls a physical body rather than answering on a screen.
Manufacturing is the other constraint. Market researcher TrendForce has pointed to the supply of joints and hands, along with sensors, as the parts that must scale for volumes to rise, and argued that makers now have to show real value to customers rather than impressive balance or demos. Price is part of that too: Morgan Stanley put a typical humanoid at around $200,000 in 2024 and expects roughly $150,000 by 2028.
That is why the first buyers are businesses. Berg Insight lists car and electronics factories, aerospace, logistics and energy alongside caregiving and elderly care, while Goldman expects warehouses and car plants to lead. Homes come later, if at all, in most of these forecasts.
For the IoT industry, the lesson is that a humanoid is a connected device first and a robot second: it has to be tracked, updated, secured and supported for a decade, often across several countries. We covered what happens when AI starts acting on devices in where device safety should begin. Whether 26 million arrive a year or far fewer, the open question is who will run the networks and the fleet software that keep them working, and how much of that value the robot makers keep for themselves.
