How many humanoids will sell in 2026: the counted answer, the forecasts, and the gap

About 19,100 humanoids shipped worldwide in the first half of 2026. That is the only counted number for the year so far. Smart Analytics Global published the count in August 2026. The forecasts for the full year run from about 60,000 to 90,000 units. They come from SAG, from Goldman Sachs in September 2026, and from Bank of America in March 2026. Nobody knows the 2026 total yet. Anyone who quotes one is extrapolating, not counting. This page keeps the counts, the forecasts, and the gap between them separate.

The counted numbers, source by source

Start with what someone actually counted. Each row below is a counted figure. It carries the counter, the date, and what the counter includes.

The 2025 pair is the important one. Omdia's ranking, published through AgiBot's January 2026 release, puts 2025 shipments at about 13,000 units. AgiBot shipped more than 5,100 of them, a 39 percent share. The International Federation of Robotics, on September 30, 2026, counted about 7,000 full-size humanoids sold for commercial and professional use. Both counts are correct. They count different things. The next section explains exactly what.

For 2026, SAG's August 10, 2026 release counted about 19,100 units in the first half. That is up 272 percent from 5,100 a year earlier. AgiBot shipped about 8,400 of them and took 44 percent of the world market, passing Unitree at about 5,900. Galbot followed at about 900, then UBTECH at about 700 and Leju at about 600. Two makers, AgiBot and Unitree, built about three quarters of all humanoids shipped on Earth in those six months.

What each counter actually counts

The spread between 7,000 and 13,000 for the same year is not an error. It is scope.

The IFR counts robots sold for commercial and professional use. Reuters reported the IFR definition on September 21, 2026. A humanoid must look human. It must work on its own in places built for people. Legs are not required. The IFR also excludes consumer and military humanoids. Omdia counts shipments of all humanoids. SAG counts what it calls functional embodied humanoids. It excludes toy-class lines such as Aelos and Roban by name.

Scope explains the rest of the noise too. A robot sold to a university lab counts as one shipment. So does a robot dancing at a product event. So does a robot working a ten hour factory shift. They are not the same kind of progress. Each tracker listed above books them the same way. The IFR said it plainly in September 2026. Most current applications remain specialized and often require human teleoperation. Safety standards, high training and maintenance costs, and the lack of a strong business case, the IFR wrote, continue to limit widespread use.

Reuters, quoting the IFR secretary general on September 21, 2026, added the detail that matters most here. Many humanoids sold in 2025 were not doing productive work. Research institutions bought them. So did companies using them to generate training data for AI models. Carmakers, the expected first buyers, run pilots with single digit or sometimes double digit robot counts.

The honest reading of the counted layer is short. About 19,100 shipped in half a year. Growth ran at almost four times. The buyers are still labs, stages, and data teams more than shift labor.

The 2026 forecast set, named and dated

These are forecasts, not counts. Each row carries the forecaster and the date. Most of these numbers have already been revised at least once.

Two of these rows deserve a closer look. Goldman Sachs, in its September 2026 revision, now expects 75,000 units in 2026. That sits inside a much larger long-range view. Bank of America's Physical AI research, dated March 2026, expects 90,000 in 2026 and 1.2 million a year by 2030. The J.P. Morgan Private Bank view, published August 14, 2026, says the three leading Chinese makers are on track to produce more than 100,000 units this year. That is about five times their 2025 output. Note the quantity there. It counts production by three makers, not global shipments by everyone. It is a different thing again.

The second-half math

Nobody has published a counted full-year 2026 number. But the counted 19,100 puts a hard test on each forecast. This arithmetic is ours. It uses SAG's count as the base.

Hardware makers usually ship more in a second half than a first. Some tilt is normal. The capacity is real too. UBTECH's Liuzhou plant is built for 10,000 units a year. Figure's first high-volume line is built for up to 12,000 humanoids a year. The company reported its 1,000th Figure 03 built on July 23, 2026. XPeng, as reported, targets mass production of its IRON humanoid by the end of 2026.

Two caveats sharpen that table, and they change how to read it. First, the base. Every multiple above sits on SAG's count alone. The 2025 rows showed why that matters. Two honest counters put the same year at 7,000 and 13,000. Count the first half on a wider or narrower scope and every share and every multiple moves with it. Second, the geography. SAG puts Chinese makers above 97 percent of H1 shipments. So the second-half ask is, in practice, a Chinese capacity ramp. The one named US line, Figure, is built for up to 12,000 humanoids a year. That full-year ceiling is under a third of the smallest second-half gap in the table. That arithmetic is ours too.

Still, the shape of the ask is blunt. Hitting 60,000 means doubling the first half inside six months. Hitting 90,000 means close to quadrupling it. The year could also land near 40,000 to 50,000. Then each bank number in the table was high. The counted numbers will say so in early 2027, when SAG and Omdia publish their full-year counts.

Why 2026 is the test year

The growth is real. SAG's counts show 5,100 units in the first half of 2025 becoming 19,100 a year later. Omdia's 2025 count of 13,000 stands against an implied 2,200 or so for 2024. Those are steep curves by any hardware standard.

What makes 2026 a test rather than a victory lap is the distance to the bank curves. The banks need the second half to run at two to four times the first. The IFR's September 2026 reading of the same market is cautious. Teleoperation is still common. There is no strong business case in industrial settings yet. The J.P. Morgan Private Bank, in the same August 2026 piece that flagged the 100,000-unit Chinese ramp, also wrote this: the gap between a viral demo reel and a truly useful general-purpose robot remains quite wide.

Both things are true at once. Shipments are climbing fast. Almost none of the shipped units are doing shift work yet. That is what a test year looks like from inside.

The long-range spread, labeled

The 2026 disagreement is small next to the 2030 one. Take the same year, 2030. Omdia said 38,000 units in July 2024. Bank of America said 1.2 million in March 2026. That is a spread of more than 30 times between named houses. Goldman Sachs sits between them at 890,000 for 2030. That figure is about 23 times Omdia's, and the bank raised it from 256,000 in its September 2026 revision. Omdia's own words in that July 2024 report still read well. Humanoid robot technology remains in its infancy. Large-scale implementation is likely a few years away.

Longer out, the numbers stop being shipments at all. That is where most confusion enters.

• Morgan Stanley, May 14, 2025: about 1 billion humanoids in use by 2050. About 90 percent would do industrial and commercial work. The 5 trillion dollar market figure includes supply chains, repair, and support. Adoption, in their words, should be relatively slow until the mid-2030s.
• Bank of America, March 2026: 3 billion humanoids in operation by 2060.
• Goldman Sachs, September 2026: 6.48 million units shipped in 2035, worth 138 billion dollars.

The first two are installed base. They count machines that exist, not machines sold per year. The third is annual shipments. These are not comparable rows. Any article that stacks them in one column is mixing units.

One more ceiling check, because trillion-unit headlines circulate. No credible research house forecasts trillions of robot units. The credible unit ceilings are about 1 billion humanoids by 2050 from Morgan Stanley and 3 billion by 2060 from Bank of America. Set those against a world labor force of about 3.74 billion people, the World Bank figure for 2025. The Bank of America number works out to about 0.8 robots per working human, close to four humanoids for every five people who work. That division is ours. Morgan Stanley's 1 billion by 2050 is the one near one robot for every four workers. The trillions live in revenue forecasts, decades out, with supply chains and support folded in.

For scale, the robot population that already exists is not humanoid. The IFR counted a record 5 million industrial robots working in factories at the end of 2025. It forecasts 655,000 new installations in 2026. Humanoids, at each count above, are a rounding error on that base so far.

Where the money actually splits

If the units arrive, where does the money sit? Not all in hardware.

Goldman Sachs, in its own robotaxi work dated April 2026, splits a roughly 2 trillion dollar 2035 autonomous-vehicle market. About 300 billion dollars of it is AI software and subscriptions, a 15 percent share. Run the same share on Goldman's own 138 billion dollar humanoid market for 2035. The AI and software layer comes out near 20 billion dollars a year. Our own bottom-up estimate, built from hosted serving rates, lands in the same range. Two methods, one answer. The robot AI software layer in 2035 is a 10 to 30 billion dollar per year market. That arithmetic is ours, not a bank's. It is also why a units count matters even while the units are mostly lab machines.

The buyers tell the same story from the other end. The IFR found that many 2025 humanoids went to research institutions and data companies. The first spending wave behind these robots is training data and evaluation, not shift labor. The hardware tally and the software tally are the same market, measured at two different times.

What this means if you build robots

You are reading a market page, so here is the builder's cut. Each unit in the tables above still needs a policy, a brain that turns cameras into motor commands. It also needs the data to train that policy. That is the layer this site works in.

The 2026 buyer profile means your earliest demand is training and evaluation, not 24 hour shift work. The light entry point is real and cheap. An ACT policy trains on a used desktop card in about an hour. The tier 1 builder budget guide prices that path from zero dollars up. At the research end, a pi 0.5 class policy runs a full manipulation loop. In our own October 2026 cost runs, serving a pi 0.5 class policy for one robot working one hour cost about 6.48. The cost per robot hour page shows that math in full. The per-robot-hour glossary defines the unit.

Whether the 2026 count lands at 45,000 or 90,000, the direction is set. More machines, more data demand, and a software layer growing inside each forecast. Want to compare what hosted model routes cost as that layer grows? The live rates sit on the pricing page. You can open an account at signup. The companion pages go deeper on each layer. The 2025 actuals cover last year's counts. Why forecasts disagree takes apart the 2030 spread.

Frequently asked questions

Will 2026 hit the bank forecasts? On the counted pace, hitting them is hard. The first half delivered 19,100. SAG's own 60,000 forecast needs a second half at 2.1 times that. Bank of America's 90,000 needs 3.7 times. Second-half weighting is normal in hardware. Those multiplies are still steep. The counted answer arrives in early 2027.

How many humanoid robots exist right now? Counted shipments give a floor. About 2,200 shipped in 2024. About 13,000 in 2025 on the widest scope. Another 19,100 in the first half of 2026. That puts the world past 34,000 shipped humanoids in two and a half years. Many of those machines are lab, display, or data collection units. "Exists" is a bigger number than "works".

Why do the 2025 counts disagree, 7,000 versus 13,000? Scope. The IFR counts full-size humanoids sold for commercial and professional use. It excludes consumer and military units. Omdia counts all shipments. Both are honest counts of different things. Read the scope line before quoting either.

Is China really 97 percent of this market? For shipments in the first half of 2026, SAG puts Chinese makers above 97 percent of global shipments. It puts China above 85 percent of demand. J.P. Morgan's Private Bank puts China at about 75 percent of the humanoid market in its August 2026 analysis. Any 2026 forecast is, in practice, a forecast of the Chinese market.

When does the real 2026 number land? Early 2027, when SAG and Omdia publish their full-year counts. Until then, treat any full-year 2026 figure as a projection. That includes each one above 19,100 in this page's forecast table.

Last verified: 2026-10-26