Humanoid robot forecasts compared: Goldman’s 75,000 for 2026 vs 19,100 shipped in H1

As of September 28, 2026, bank forecasts for 2026 humanoid shipments range from about 60,000 to 90,000 units, while the only half-year count so far, from Smart Analytics Global (SAG), puts first-half 2026 shipments at about 19,100. Long-range forecasts diverge far more: Goldman Sachs now sees 6.5 million units and a $138 billion market in 2035, Morgan Stanley about 1 billion humanoids in use and a $5 trillion market by 2050, and RBC a $9 trillion addressable market by 2050.

Why it matters: these numbers underpin valuations, supplier capex and state funding. The gap between a 2025 base of 7,000 to 13,000 units and multi-million-unit forecasts is where most of the sector’s capital risk sits.

How many humanoid robots were actually shipped in 2025 and 2026?

Between 7,000 and 13,000 in 2025, depending on who counts and what they count. Omdia’s January 9, 2026 report put 2025 shipments at about 13,000, led by AgiBot (5,168 units), Unitree (4,200) and UBTech (1,000). The International Federation of Robotics put 2025 sales at about 7,000, but its scope is narrower: industrial and professional service robots only, excluding consumer and military humanoids (our IFR report). The IFR also said many of those units went to research institutions and to companies collecting training data.

For 2026, SAG reported first-half shipments of about 19,100, up 272% on a year earlier, with AgiBot at about 8,400 and Unitree at about 5,900. SAG says industrial and commercial uses took more than 70% of first-half shipments, up from about 50% a year earlier, and forecasts about 60,000 for the full year. Behind the two leaders the counts drop sharply: Galbot at about 900, UBTech about 700 and Leju about 600. SAG puts Chinese vendors at more than 97% of global shipments and China at more than 85% of demand, so any 2026 forecast is in practice a forecast of the Chinese market. Company-by-company totals are in our humanoid shipments tracker.

What do Goldman Sachs, Morgan Stanley and other banks forecast for humanoid robots?

The table lists the main published forecasts as of September 28, 2026. All are projections; unit figures mean annual shipments unless marked as units in use.

Forecaster	Date of forecast	Horizon	Units	Market value	Notes
Goldman Sachs	Feb 27, 2024	2030 / 2035	250,000+ / 1.4M shipments	$38bn (2035)	Superseded
Goldman Sachs	Sep 2026 (reported Sep 13–14)	2026 / 2030 / 2035	75,000 / 890,000 / 6.48M	$138bn (2035)	Figures from published coverage of client report
Bank of America	Cited by Reuters, Sep 21, 2026	2026 / 2030	90,000 / 1.2M shipments	Not given	Highest 2026 figure
SAG (research firm)	Aug 2026	2026 / 2030	~60,000 / 500,000	~$1.6bn revenue (2026)	Based on H1 tracking
Barclays	Jan 14, 2026	2035	Not given	$200bn (optimistic scenario)	Market put at $2–3bn today
Barclays	Sep 18, 2026 (as reported)	~2035	Not given	Not given	Mass commercialisation of general-purpose humanoids moved to around 2035, from 2030
UBS	Jun 2025	2035 / 2050	2M / 300M	$30–50bn / $1.4–1.7T	Adoption could take “over five years”
Citi	Nov 4, 2024	2050	Almost 650M in use	$7T	Citi GPS report
Morgan Stanley	May 14, 2025	2050	~1B in use (~90% industrial/commercial)	$5T incl. supply chain, repair, maintenance	Slow adoption until mid-2030s
RBC Capital Markets	Mar 27, 2026	2050	Not given	$9T addressable market	China ~61%; households ~33%

Two leads we checked could not be pinned to a primary source and are not in the table: a precise “648 million” Citi figure (Citi’s own page says “almost 650m”) and any Morgan Stanley shipment forecast for 2026. Goldman’s September 2026 figures come from 24/7 Wall St. and Humanoid.guide coverage of a client report (our Goldman report).

Are the 2026 humanoid forecasts on track?

Not on a straight-line basis. At 19,100 units, the first half of 2026 delivered about a quarter of Goldman’s full-year figure. Hitting any of the three 2026 forecasts requires the second half to be at least double the first.

2026 forecast	Full-year units	H1 actual as share	H2 units needed	H2 vs H1
SAG	~60,000	32%	~40,900	2.1×
Goldman Sachs	75,000	25%	~55,900	2.9×
Bank of America	90,000	21%	~70,900	3.7×

Calculations are ours, using SAG’s 19,100 first-half figure. Second-half weighting is normal for hardware makers, and capacity is arriving: UBTech’s Liuzhou plant is designed for 10,000 units a year and XPeng targets IRON mass production by the end of 2026. But a doubling of output within six months is a demanding test, and the higher bank figures need close to a quadrupling.

Why do humanoid robot forecasts differ so much?

They measure different things. Some count annual shipments, some units in use, and some the value of a whole ecosystem: Morgan Stanley’s $5 trillion includes supply chains, repair, maintenance and support, and RBC’s $9 trillion is an addressable market, not expected revenue. Price assumptions drive the rest. Morgan Stanley assumes unit prices fall from about $200,000 in 2024 to about $50,000 in high-income countries by 2050; RBC assumes an average near $25,000. Goldman’s 2035 revision raised units about 4.7 times but market value only about 3.6 times, implying a lower price per robot than its 2024 model.

The implied prices are worth checking against each other. SAG’s forecast of about $1.6 billion in 2026 revenue on about 60,000 units works out to roughly $27,000 per robot. Goldman’s $138 billion on 6.48 million units in 2035 implies about $21,000. Both sit far below the $200,000 Morgan Stanley estimated for a humanoid in a high-income country in 2024, which reflects how much of today’s volume is low-priced Chinese hardware sold to labs and exhibitors. Our per-unit figures are simple division, not the forecasters’ own numbers.

Scope explains the rest of the spread. Bank of America’s 90,000 for 2026 is about 13 times the IFR’s 2025 sales count but only about seven times Omdia’s shipment count for the same year. Neither base is wrong: the IFR excludes consumer and military units, while Omdia counts all shipments. A forecast quoted without its base year and scope cannot be compared with anything.

The direction of revisions has also split. Goldman has upgraded twice in about two and a half years. Barclays, by contrast, sized a $200 billion 2035 market in January 2026 and in a September 18 note moved its expected date for mass commercialisation of general-purpose humanoids to around 2035, from 2030, arguing that intelligence rather than hardware is the bottleneck (our Barclays report). Near-term demand also looks different from the long-range story: much of today’s humanoid revenue comes from research, education and display, not factory work (our analysis of where physical AI revenue is booked).

The useful checkpoints are close. SAG’s and Omdia’s full-year 2026 counts, due in early 2027, will show whether the year landed nearer 40,000 or 75,000. Until then, treat any 2030-plus figure as a scenario, and the 2026 figures as the only ones that can yet be tested.