Humanoid Robots: Announced vs. Shipped Units, 2025–2026

Key Takeaways

• The same half-year has three different counts.For H1 2026, SAG puts global shipments at about 19,100, Counterpoint above 22,000 and IDC at nearly 25,000, a spread of about 5,900 units that reflects method rather than a shortfall. For 2025, IFR's narrower tally (about 7,000) sits against Omdia's 13,318.
• Growth is steep inside every tracker.Year-on-year H1 2026 growth is reported at 272% (SAG), nearly 300% (Counterpoint) and 432.1% (IDC). Growth rates are comparable within a source; absolute totals are not comparable across sources.
• China holds the disclosed volume.SAG attributes more than 97% of H1 2026 shipments to Chinese vendors. AgiBot ranks first in all three trackers, with roughly 8,400 to 9,700 units, itself a 1,300-unit range for one company.
• Capacity announcements outrun disclosed units.Unitree has committed to 75,000 humanoids a year of capacity and Figure's BotQ is designed for up to 12,000, against tracker estimates of about 5,000 to 5,900 H1 Unitree shipments and more than 350 Figure 03 robots delivered by April 2026. The metrics differ, so NextMSC calculates no gap.
• Full-year forecasts need a steep second half.Counterpoint's 50,000-plus implies H2 shipments of about 1.3 times H1; SAG's roughly 60,000 implies about 2.1 times; Bank of America's 90,000 implies 2.6 to 3.7 times, depending on the H1 tracker used.

2026 Market Pulse Metrics

At-a-glance benchmarks, each attributed to its publisher and kept in that publisher's own definition.

"~" marks an approximation and ">" a lower bound as stated by the publisher. Omdia's AgiBot share is 38.8% on NextMSC's arithmetic (5,168 ÷ 13,318); coverage rounds it to "nearly 38%" or 39%.

Executive Summary

The state of the count

Between 2025 and mid-2026, humanoid shipment estimates rose several-fold in every dataset NextMSC reviewed, yet no two datasets agree on the base. Omdia counted 13,318 shipments for 2025. The International Federation of Robotics (IFR), counting only industrial and professional-service sales, reported about 7,000, so Omdia's total is roughly 1.9 times IFR's. For the first half of 2026 the range is 19,100 to nearly 25,000. Trackers differ on what qualifies as a humanoid, whether they count shipments, sales or installations, and which vendors they can observe. NextMSC therefore reports each estimate separately.

For scale, the same IFR release cites about 542,000 conventional industrial robots installed in 2024 and about 199,000 professional service robots sold that year.

Humanoids are still a small fraction of the broader AI robots picture, and IFR's own secretary general described them that way.

The measurement ladder

The more consequential problem sits upstream of the trackers. Company announcements are quoted in four different currencies (design capacity, cumulative production, orders and shipments) and headlines routinely set one against another. AgiBot's 20,000 cumulative robots produced by September 2026, for example, spans several robot forms, and IDC's more than 8,600 AgiBot H1 shipments counts humanoids only over six months. The two numbers answer different questions. NextMSC's reading is that public data is dense at the top of the ladder (capacity and orders) and thin at the bottom (shipments to paying users and sustained deployment).

Five Metrics That Get Mistaken for Shipments

Each row is a different measurement with a verified example from this report's source set.

Source: Company releases and filings, TrendForce, Counterpoint Research, Reuters and NextMSC analysis.

Same Half-Year, Three Counts

Three research firms measured the first half of 2026 and returned three answers (Figure 1). IDC's "nearly 25,000" is roughly 31% above SAG's 19,100, a difference of about 5,900 units. Even the leader's volume depends on the tracker: AgiBot's H1 shipments run from about 8,400 (SAG) to about 9,700 (Counterpoint), a 1,300-unit, or about 15%, difference for one company.

Baselines differ before any forecast is applied. Counterpoint's 50,000-plus 2026 forecast and its stated 210% growth together imply a 2025 base near 16,100 units (50,000 ÷ 3.1, NextMSC arithmetic), above Omdia's 13,318 and more than double IFR's narrower figure.

† IDC's Unitree figure is as reported in secondary coverage of IDC's vendor table; IDC's own table was not accessed. ‡ Counterpoint's vendor chart was not readable as text; its Unitree unit count could not be confirmed and is omitted. Source: IDC, Counterpoint Research and Smart Analytics Global, as published or reported.

Who is receiving the robots?

Application mix is as contested as the totals. SAG says industrial and commercial uses made up more than 70% of H1 2026 shipments, up from about 50% a year earlier. Counterpoint says entertainment and performance plus data production and research together remained above 60% of shipments, edging down. IDC, as reported by Humanoids Daily, puts research and education, performances and displays, and government data centers at 69% of H1 shipments, down from 83.8% across 2025. The categories overlap and are defined differently, so NextMSC does not reconcile them. It does read them together as evidence that a large share of shipped units still go to research, data collection and demonstration, not to production lines. Reuters' reporting on the IFR data, as summarized by TechRepublic, points the same way: IFR's secretary general said carmakers are running pilots with single-digit or low double-digit robot counts per plant.

Announced Capacity Versus Disclosed Units

China: volume is disclosed, definitions are not aligned

AgiBot, Unitree, UBTECH, Leju and Galbot account for most disclosed volume. Counterpoint reports Leju at 650 units (2.9% share) and Galbot above 1,100 (about 5%) for H1 2026. Disclosure quality still varies. Unitree stated in January 2026 that it delivered more than 5,500 humanoids to end customers in 2025 and produced more than 6,500, and that these figures exclude its wheeled robots. That delivery figure exceeds Omdia's 4,200 estimate by at least 1,300 units, about 31%, and Unitree itself advises against combining counts across robot types. Both company numbers are thresholds, not exact counts. NextMSC reads the episode as evidence that vendor-level reporting is overtaking tracker estimates as the better unit source.

UBTECH offers the clearest side-by-side of target and filing. In November 2025 it targeted 500 Walker S2 deliveries by year-end and annual capacity of 5,000 units in 2026 and 10,000 in 2027. Its H1 2026 results report 921 full-size humanoid units sold, and coverage of those results cites a 2026 production target of 10,000 full-size humanoids. Sales and production targets are different metrics, so 921 is shown as context (9.2% of 10,000), not as a miss. Gasgoo's coverage of the same filing reports 16,123 units across all humanoid categories, a reminder of how far the word "humanoid" can stretch.

United States and Europe: capacity, orders and pilots lead

Figure said on 29 April 2026 that it had delivered more than 350 Figure 03 robots and demonstrated a cycle time of one robot per hour, but also stated that robots shipped from BotQ are allocated to internal research, data collection, household-task work and commercial use-case development. Agility's more than $300 million of multi-year orders relate to Digit 5, with early access reported for the first half of 2027, and include a 1,000-robot, three-year robots-as-a-service contract with a single customer. 1X says first-year capacity of more than 10,000 NEO units sold out in five days, yet its April 2026 factory output was going to its own R&D lab and home testing, with first customer shipments planned for 2026. Tesla's Q2 2026 materials commit to Optimus production in 2026, with Musk describing the early part of the ramp as flat and long. Hyundai's 30,000-robot system is a 2028 ambition.

Announced-vs-Shipped Evidence Matrix

NextMSC compares each company's announced figure with its latest disclosed unit evidence, and reports a numeric gap only where company, model scope, period and metric all match. None currently do.

Source: Company releases, filings and earnings materials; TrendForce; Omdia, IDC, Counterpoint Research and SAG as published or reported; The Robot Report and TechRepublic on Agility's S-4; NextMSC analysis. Arithmetic marked as illustrative is NextMSC's.

Can the 2026 Forecasts Be Met?

Forecasts are the loudest announcements of all. Applying each forecast to the first-half count shows what it demands from the second half (Figure 3). Forecasts are stated as "exceed" or "approach", so the arithmetic uses the stated thresholds and is approximate. Only Counterpoint and SAG pair a forecast with an H1 count from the same publisher; Bank of America's figure is tested against two other trackers and is therefore cross-source.

Source: Counterpoint Research, Smart Analytics Global, IDC, Bank of America Global Research (via Reuters) and NextMSC analysis. Implied H2 = full-year forecast minus H1 comparator.

NextMSC's reading: Counterpoint's forecast needs modest sequential growth, while Bank of America's would need H2 shipments to be a multiple of everything shipped in H1. That is a step-change, and one that depends on orders converting into deliveries within months. The H1 composition above, with research and demonstration still a large share, is a reason to treat the higher forecasts as a test of conversion, not a baseline. Anyone sizing the wider AI robots opportunity should apply the same discipline to any market-size figure.

Dataset Explorer: Verified Data Points

Every announcement, milestone and unit count used in this report, one row per data point, with its ladder step, period and source tier (A = published by the originator, B = reported by a third party).

n/d = no verified public count. Source: Company releases and filings, TrendForce, Counterpoint Research, Omdia, IDC, SAG, IFR, Reuters and NextMSC analysis.

Methodology & Confidence Key

This report draws a hard line between published data, reported data and NextMSC calculations. It records numeric claims found in public sources, preserves each source's definition, and never averages across publishers. Data cut-off is 9 October 2026; full-year 2026 actuals were not available at cut-off.

Definitions used

• Capacity: designed or targeted annual output of a plant or line. Not output.
• Cumulative production: units rolled off a line since launch, as the company defines its portfolio.
• Shipments / sales: units counted by a tracker or reported by a company as sold or delivered. Scope differs by source.
• Orders / pre-orders: commitments that may be conditional, multi-year and not yet delivered.
• N/C: no defensible same-scope, same-period arithmetic is possible from public disclosures.

Conflicts identified and how they were handled

• AgiBot 2025 share: SCMP reports "nearly 38%" and AgiBot's own release 39%. NextMSC's calculation from Omdia's unit figures is 38.8%.
• UBTECH 2026 target: the November 2025 release states 5,000 units of annual capacity by 2026; coverage of the August 2026 interim results cites a 10,000-unit 2026 production target. Both are shown; the second was not confirmed in UBTECH's own wording.
• AgiBot March 2026 milestone: TrendForce calls the 10,000th unit a general-purpose embodied robot; some coverage calls it a humanoid. TrendForce's wording is used.
• Application mix: SAG, Counterpoint and IDC use different categories and point in different directions. They are shown side by side without reconciliation.
• Tesla call date: reporting indicates Q2 2026 results were released in July 2026, while some transcript pages carry later dates. The report cites "July 2026".
• Excluded: Counterpoint's separate 2025 installations estimate (publication date could not be confirmed) and numeric Tesla Optimus targets that conflict across secondary sources.

What remains unknown

Verified public unit totals were not found for Tesla, Figure's external customers, Agility, 1X or Hyundai/Boston Dynamics. Absence of disclosure is not evidence that zero units shipped. No dollar order values were converted into unit counts.

Planned refresh: quarterly, following new tracker releases and company results. Corrections and source challenges: contact NMSC research.

Outlook: What to Watch Next (NextMSC Analysis)

The mechanisms behind scale are visible: TrendForce expects China's 2026 humanoid output to grow by up to 94% and expects Unitree's capacity commitment to drive upstream expansion in joint modules, dexterous hands and sensors. Input supply is a further variable; NextMSC's analysis of how AI demand is tightening memory supply shows how semiconductor inputs can reprice quickly. The signals below are what NextMSC would monitor. They are monitoring points, not unit forecasts.

Second-half tracker releases

Full-year 2026 actuals from Omdia, Counterpoint, IDC and SAG will test whether H1's spread narrows or widens, and whether H2 reaches the multiples in Figure 3.

First disclosed Western unit counts

1X's first customer shipments (planned for 2026), Tesla's Optimus production start and any Figure customer-delivery disclosure would move several rows of the matrix from "not disclosed" to measurable. Agility's Digit 5 early access is reported for H1 2027.

Reporting convergence

Unitree's advice not to combine counts across robot types points to the core fix: shipments by robot form, application and customer type on a stated definition. Audited disclosures from vendors in the listing process, including Unitree's STAR Market application noted by TrendForce, may provide the first audited unit data.

Implications for stakeholders

• Buyers and procurement: ask vendors which step of the ladder a quoted number refers to, and tie delivery schedules to shipments, not capacity.
• Investors and analysts: reconcile each claim to a metric before comparing companies; treat conditional multi-year orders as backlog, not revenue.
• Component suppliers: plan on shipment run-rates and application mix, not nameplate capacity.

Forecasts and capacity figures in this report are announced plans or third-party projections. Modeled and calculated values are NextMSC analysis and should not be cited as official statistics.

Conclusion: Measure Conversion, Not Announcements

The evidence supports two statements that are easy to confuse. Humanoid shipments are rising quickly: every tracker NextMSC reviewed reports H1 2026 year-on-year growth of at least 272%, and company disclosures such as Unitree's more than 5,500 deliveries in 2025 confirm that real units are moving. Yet announced capacity and orders are far larger than disclosed units, and the trackers disagree on the base by thousands of units.

NextMSC's reading is that the sector is real but early. Disclosed volume is concentrated in China and in channels that include research, education and demonstration, while US and European players lead with capacity, orders and pilots rather than delivered counts. The 2026 forecasts, from more than 50,000 to 90,000, require second-half shipments of roughly 1.3 to 3.7 times the first half.

For buyers, investors and suppliers, the practical step is the same: identify which rung of the ladder a number describes before using it. The next tracker releases and the first disclosed Western unit counts will show whether announced capacity is converting into shipments. NextMSC will refresh this analysis quarterly.