It was reported on August 31, 2026, that Goldman Sachs had revised its 2035 humanoid robot market forecast to $138 billion and 6.48 million units. This came just two and a half years after the firm raised the figure to $38 billion and 1.4 million units in February 2024. Meanwhile, a parts-cost estimate published by the Bank of America Institute in April 2025 puts the bill of materials (BOM) per unit at $35,000 by the end of 2025, dropping to only $13,000–$17,000 by 2030–2035 even as mass production ramps up. While the dollar forecast moved 3.6x in two and a half years, the cost projection only assumes it will fall by less than half over five years. Closing this gap in speed isn't happening on the forecasting side — it depends entirely on how far the unit price of the roughly 30 actuators packed into each robot can fall.
A 6x Jump, Then Another 3.6x in Two and a Half Years
Goldman Sachs Research's November 15, 2022 report, "Humanoid Robots: Sooner Than You Might Think," contained no dollar figure for market size. What it did say was that "in the history of humanoid robot development, no robot has yet achieved commercial success," along with a projected timeline in which economic viability might emerge for factory applications between 2025–2028 and for consumer applications between 2030–2035. The report also noted that recouping investment within two years would require cutting production costs by 15–20% annually.
Fifteen months later, the firm put a number on it. In its February 27, 2024 revision, Goldman set the 2035 total addressable market (TAM) at $38 billion — more than six times the $6 billion figure it cited as its own prior estimate. Unit shipments were pegged at 1.4 million, four times the earlier projection. The report was led by Jacqueline Du, who heads the firm's China Industrial Technology research. The reason given for the revision was a single line: "AI progress has surprised us the most." There was no mention of actual deployment results.
Costs had moved too. The same report noted that per-unit manufacturing costs had fallen from the prior year's range of $50,000–$250,000 (from budget to state-of-the-art models) to $30,000–$150,000. Against the firm's own assumption of a 15–20% annual decline, the actual drop was 40%. And the reason Goldman itself gave for this decline wasn't a design breakthrough but wider access to cheaper components and more suppliers. In other words, sourcing — not engineering — was what drove the cost down.
Then, at the end of August 2026, the firm reportedly rewrote the same 2035 forecast once more. Market size: $138 billion. Unit shipments: 6.48 million (likely annual shipments, though Goldman's own classification wasn't specified in the reporting). The 2030 figure was reportedly raised from 256,000 to 890,000 units, and 2026 from 51,000 to 75,000.
Goldman Sachs has now raised its outlook for the same 2035 market twice — from $6 billion to $38 billion to $138 billion. Unit shipment projections have climbed 4.6x, from 1.4 million to 6.48 million. The scale of these forecast revisions far exceeds the actual pace of BOM cost declines. That said, the August 2026 figures haven't been confirmed on Goldman's official page, and reporting doesn't clarify whether the 6.48 million figure for 2035 represents annual shipments or a cumulative total. Even the original $6 billion figure was one that the February 2024 report retroactively cited as its own prior estimate.
Such large swings in forecasts stem from how TAM figures are constructed. These numbers are typically top-down estimates, calculated by multiplying total labor market size by a penetration rate — and moving that multiplier by even one percentage point can shift the result by hundreds of billions of dollars. CleanTechnica's Michael Barnard has offered a conditional calculation: applied to a $20 trillion global labor market, a 1% penetration rate yields $200 billion, while 5% yields $1 trillion (he is not forecasting the penetration rate itself at 1–5%). If one percentage point of assumption equals $200 billion, revising the forecast is hardly difficult.
Cost estimates, by contrast, are built bottom-up. The Bank of America Institute's "Humanoid robots 101" (April 29, 2025) assumes Chinese-made components and puts the end-2025 BOM at $35,000, projecting it will fall to $13,000–$17,000 by 2030–2035 — an implied annual decline rate of roughly 14%. Extrapolating that 14% rate directly, it would take about 5.6 years for $35,000 to reach the median target of $15,000 — landing around 2031. Over the same period in which the dollar forecast moved 3.6x, the cost estimate — even if it proceeds exactly as projected — would fall by less than 30%.
The Job of Turning a $930 Joint Into a $240 Joint
According to the industry analysis site GrabaRobot, the parts cost of Tesla's Optimus Gen-2, excluding software, is estimated at $50,000–$60,000 per unit. Per the same article, actuators account for the largest share of that cost at roughly 56%, followed by hands at 17.2% ($9,500), then the waist and pelvis at 14.2%, thighs at 13.2%, shins at 13.2%, and feet at 12.2%. This breakdown is attributed to Morgan Stanley, but it hasn't been directly verified against the original source and should be treated as a secondhand figure.
The percentage itself varies by source. Estimates range from 35–40% (TechTimes) to roughly 56% (Morgan Stanley's Optimus Gen-2 estimate) to 40–55% (independent estimates) — scattered enough that no single figure can be stated definitively. What's certain is only that actuators occupy somewhere within a 40–60% range, depending on the estimate.
The Bank of America Institute's model offers a concrete count: 16 rotary actuators and 14 linear actuators per unit, totaling 30, with hand dexterity set at 6 degrees of freedom. The institute states that a 6-DOF hand design can cover 60–70% of the functionality of a human hand (which has 27 degrees of freedom), and its cost estimates assume a hand at this level.
Applying the percentage and count yields a per-unit price. Using a model where actuators account for 56% of BOM across 30 units, the parts cost per joint comes to $930–$1,120. To bring the BOM down to $13,000–$17,000 while keeping the same ratio and count, each joint would need to fall to $240–$320 — a roughly 73% reduction. The math is straightforward: $50,000 × 0.56 ÷ 30 units = $933; $60,000 × 0.56 ÷ 30 units = $1,120. On the target side: $13,000 × 0.56 ÷ 30 units = $243; $17,000 × 0.56 ÷ 30 units = $317. Comparing the medians, the required reduction rate is 73%.
This calculation comes with caveats. The BOM total and ratio are Morgan Stanley's estimate for a single model — Optimus Gen-2 — not an industry average, while the count of 30 units comes from a Bank of America Institute model with a different publisher and year. The target range itself is an April 2025 forecast, not a realized price. Nor is there any guarantee the actuator ratio will remain at 56% going forward — if other cost categories fall first, the ratio could actually rise. Still, the order of magnitude — roughly $700 per joint that needs to be cut — holds regardless.
So where does that $700 come from? Here's where the difference from semiconductors matters. With semiconductors, each generation packs more transistors into the same area, causing per-function unit costs to fall in a step-like pattern.
Precision reducers work the opposite way — the process of grinding, measuring, and assembling gear surfaces doesn't shrink. Whether it's a wave gear device or a planetary roller screw, what determines unit price is the precision of the machining equipment and the number of process steps — and that doesn't automatically shrink just because production volume increases. What falls is yield rate, changeover time, in-house manufacturing that eliminates middleman margins, and material procurement terms. In other words, this follows a learning curve, not an exponential curve.
Looking at current unit prices shows just how far this distance really is. According to the same GrabaRobot article, the planetary roller screws used in leg linear joints cost $1,350–$2,700 each, with roughly 12 units per full-size robot (neither the count nor the price, including the publisher, has been verified against the original source). A single screw accounts for roughly 10% of the entire target BOM. The same goes for hands — if a 6-DOF hand set costs $9,500, that alone exceeds half of the $13,000–$17,000 target. The closer a design gets to matching human dexterity, the further it drifts from the target.
A $13,500 Machine, and Machines With No Price at All
Unitree's official store sells the G1 for $13,500 (excluding tax and shipping). Standing 1,320mm tall, weighing about 35kg, with 23 degrees of freedom and roughly two hours of battery life. Looking at this single model in isolation, it appears the mass-production BOM target of $13,000–$17,000 has already been achieved.
Western commercial machines, by contrast, don't publish prices at all. Neither Boston Dynamics' electric Atlas nor Figure 02 has an official price, and the estimates that circulate don't agree with each other. Agility Robotics' Digit is largely offered through RaaS (Robotics as a Service), and the company's June 2026 investor materials cite an assumed price of roughly $8,500 per unit per month, or about $100,000 annually. The frequently cited "$30 per hour" figure doesn't refer to Digit's service fee at all — it's the fully burdened human labor cost (about $30.50/hour) used as a comparison point.
On the consumer side, 1X Technologies' NEO is priced at $20,000 outright or $499 per month ($200 reservation deposit), with shipments to US households beginning in 2026. Tesla's Optimus target of $20,000–$30,000 is Elon Musk's long-term mass-production price goal — as of mid-2026, there are no orders, no official price, and no shipments to paying customers. Target prices, quoted prices, and market prices are all being discussed side by side, using the same "dollar" unit as if they were comparable.
When average selling prices fall, there are two distinct paths behind it. One is that the cost of producing the same configuration of machine actually falls. The other is that cheaper products sell more, shifting the product mix.
Unitree's average selling price for humanoid robots dropped from 593,000 yuan in 2023 to 168,000 yuan in the January–September period of 2025. The company manufactures its own motors, reducers, joint modules, and servo drivers in-house, so it clearly captures the first effect. At the same time, its price range has widened — from the roughly $4,900 R1 to the developer-oriented EDU version at $43,900–$73,900 — which mixes in the second effect. Reading the average figure alone as "hardware got cheaper" risks conflating the two.
The breakdown by application helps distinguish between them. Counterpoint Research's August 19, 2026 report shows global shipments of 22,000-plus units in H1 2026, up roughly 300% year over year, with full-year shipments expected to exceed 50,000 units (up 210%). Market share: AgiBot at 43%, Unitree at 31%. But by application, entertainment/demonstration and data generation/research together account for over 60%, while intelligent manufacturing is only 13% and warehouse logistics just 5%. Shipment volumes are surging, but most of those units aren't robots working in factories or warehouses.
In other words, what China's mass production has solved is the problem of making cheap robots suitable for research and demos. Selling a 23-DOF, two-hour-runtime machine for $13,500 is not the same problem as selling a finished robot that runs eight hours a day in a factory, with a decade of spare-parts availability, at that same price. When IDTechEx forecasts the global average selling price falling roughly 68%, from $114,700 in 2024 to $37,000 by 2030, its baseline year is 2024 — and whether China's mass-produced units are included in that population significantly shifts the average.
84 Finished Machines, an 80% Share in Joints
A forecast released by Yano Research Institute on April 24, 2026 (based on manufacturer shipment volumes) projects the global market to reach 16,580 units in 2025, 81,690 units in 2026 (up 492.7% year over year), and 7.18 million units by 2035 — implying a CAGR of 83.5% from 2025 to 2035. In the same dataset, the Japanese market totals just 84 units in 2025, giving it less than a 1% global share. Measured in finished units, Japan is barely participating in this market at all.
Shift focus to components, and the picture reverses entirely. Harmonic Drive Systems holds an 80–90% global share of the small precision reducer (wave gear device) market, and it moved its listing to the Tokyo Stock Exchange Prime market on February 27, 2026. For the fiscal year ending March 2026, the company reported net sales of ¥59.557 billion (up 7.0% year over year) and operating profit of ¥2.567 billion. The prior year's operating profit was just ¥6 million, meaning the figure recovered by nearly four orders of magnitude in a single year. President Akira Maruyama has said, "It's rare for a component to remain in use for this long even after its foundational patents have expired."
The reason wave gear devices have held their place in robot joints lies in the structure itself. A thin metal cup is squashed into an ellipse, and shifting the meshing point with a rigid outer ring by one revolution at a time provides the reduction. Unlike arrangements using stacked gears, this produces almost no backlash, achieves a high reduction ratio of around 1:100 in a single stage, and stays lightweight. For robots that need to precisely control joint angles, a component that satisfies all three of these properties simultaneously is hard to substitute.
And humanoid robots need a lot of joints. According to the company, the number of reducers used per unit is roughly 6–10 times that of a conventional industrial robot. In this business, what matters isn't whether a single finished robot sells, but how many reducers go into that one unit.
Another Japanese company, Nabtesco, has long held ground in mid-to-large industrial robot joints with its RV reducers. The two companies dissolved their partnership in 2021; Nabtesco decided to sell its Harmonic Drive Systems shares in November 2022, and completed the full sale by June 2023 for a confirmed total of ¥37.515 billion. Japan's two major reducer makers are now approaching the new demand from humanoid robots separately.
In a December 9, 2025 assessment, TrendForce classified Japan's strategy as raising entry barriers through core components — actuators, sensors, and control systems — while the US and China push finished products into the market en masse. This doesn't mean Japan has no finished machines. Kawasaki Heavy Industries' Kaleido 9 stands 190cm tall, weighs 99kg, has 30 degrees of freedom throughout its body, and demonstrated moving a 30kg shelf at iREX 2025 in December 2025. But Japan isn't competing on unit volume.
Japan's presence on the finished-machine side has, for now, appeared as a buyer rather than a maker. Agility Robotics and Toyota Motor Manufacturing Canada announced an RaaS contract on February 19, 2026. Following a one-year pilot with three units, initial deployment will be seven units, working at plants in Cambridge and Woodstock, Ontario, to load and unload totes onto automated guided vehicles.
Tim Hollander, President of Toyota Motor Manufacturing Canada, said that after evaluating numerous robots, the company chose Digit and is pleased to improve working conditions for employees while further increasing operational efficiency. Goldman's 2026 revision reportedly also included an estimate that Toyota could produce 190,000–540,000 units by 2035 (a 3–8% global share). Japan as buyer is now beginning to be counted as Japan as producer in these outlooks too.
One clarification is worth making here: Harmonic Drive Systems' 80–90% figure refers to global share in the specific product category of wave gear devices — not reducers overall, and not robot components as a whole. It's also widely acknowledged in the industry that Chinese manufacturers are catching up fast on price. The claim that Japan monopolizes joint components isn't supported by this figure.
"Year One" Has Arrived Three Times
Chinese media's label for this industry has advanced by one year, every year. 2024 was the year of prototype releases, 2025 was "Year One of Mass Production," and 2026 is "Year One of Commercialization." Xinhuanet's January 8, 2026 article positions 2026 as the Year of Commercialization, citing High-Tech Robotics Industry Research Institute estimates of growth from 18,000 units in 2025 to 62,500 units in 2026.
Here's how the actual, contractually confirmed deployment scale looks in that "year of commercialization": Agility Robotics and Toyota Motor Manufacturing Canada, seven units. BMW's February 27, 2026 release described its Figure 02 deployment at Spartanburg as a pilot project, with just one unit deployed. That single unit supported a process producing over 30,000 X3 vehicles over 10 months in 2025, handling roughly 90,000 part movements and logging 1,250 operating hours — but Figure announced the unit's retirement on November 19, 2025. Boston Dynamics unveiled its production-version Atlas at CES 2026, but its 2026 production allocation is entirely committed to Hyundai's Robotics Metaplant Application Center and Google DeepMind, with shipments to outside customers not expected to start until early 2027.
The gap between production and actual operation is even larger. Interact Analysis data shows that global humanoid robot production exceeded 20,000 units in 2025 (a 10x jump from under 2,000 in 2024), but only about 10% — roughly 2,000 units — were actually put into real-world operation. Most of the rest went toward research, data collection, and entertainment. The firm sees the inflection point for large-scale commercialization occurring after 2032, forecasting annual shipments exceeding 700,000 units by 2035 with market revenue of roughly $15 billion.
When comparing these figures side by side, what needs aligning first is definitions, not units. For 2035 unit counts, Goldman Sachs cites 6.48 million, Yano Research Institute 7.18 million, and Morgan Stanley 13 million — but Yano's figure represents manufacturer shipments, Morgan Stanley's represents robots in operation, and reporting doesn't clarify whether Goldman's 6.48 million is annual shipments or cumulative, meaning treating it as annual shipments remains provisional. Even when numbers appear close, what they're actually counting isn't the same at all.
Shipments represent units that left the factory in a given year (a flow), while units in operation represent an accumulated stock built up from past shipments — assuming a five-year service life, the stock figure in the same market would run several times higher than the flow. Morgan Stanley's 13 million figure is reported as a forecast for units in operation, and the picture the firm actually expresses with confidence is a longer-term one: a roughly $5 trillion market and about 1 billion units in operation by 2050. Dollar figures show the same kind of mismatch. Goldman's $138 billion represents total TAM, while Interact Analysis's roughly $15 billion represents annual shipment-based market revenue — appearing to differ by 9x, but actually comparing different things entirely.
Examples of researchers grading their own forecasts against reality remain rare in this field. In a January 1, 2026 scorecard, Rodney Brooks self-graded his 2018 predictions about dexterous hands and household robots as both having failed to materialize, then added new forecasts: that practical dexterity will remain inferior to human hands even beyond 2036, and that walking humanoid robots will continue to be too dangerous to operate close to humans unless new mechanisms emerge.
Who Cuts the Cost to $240 Per Joint?
At this point, there are three paths that could realistically deliver a 73% price cut. The first is in-house component manufacturing. Behind Unitree's cutting its average selling price to less than a third in two years lies in-house production of drive components, and the same shift is now beginning among finished-robot makers too.
Figure AI's own factory, BotQ, had reached a production rate of one unit per hour by April 29, 2026 — a 24x increase achieved in under 120 days. In the process of manufacturing over 350 Figure 03 units, the company produced more than 9,000 actuators across over 10 different part numbers, and reports a first-pass yield exceeding 80% at line's end. When a finished-robot maker owns its own actuator factory, the middleman margin on each joint disappears.
The second path is production volume itself. Learning curves don't run on time — they run on cumulative production volume. Costs fall at a fixed rate every time cumulative output doubles, which means even the Bank of America Institute's "14% annual rate" is really a volume-based function translated into a yearly figure. With global 2026 shipments projected at over 50,000 units by TrendForce and 75,000 by Goldman, doubling cumulative output at this pace won't take long. But turning a $930 joint into a $240 one requires cumulative volume to multiply many times over — and the number of years that takes is determined more by absolute unit count than by the growth rate of shipments.
The third factor, semiconductors, has limited leverage. Goldman's 2026 report reportedly estimates semiconductor costs per unit at $3,000–$6,000-plus — roughly 10–20% of a $35,000 BOM. Even if semiconductor costs were halved, the BOM would only fall by $1,500–$3,000, landing at $32,000–$33,500 rather than $35,000. Cheaper computation makes the machine smarter — it doesn't make the joints cheaper.
A regulatory line has now been drawn across this landscape. On July 28, 2026, the FCC added "foreign-produced advanced robotic devices" — including autonomous mobile robots, humanoids, and quadrupeds — to its Covered List via public notice DA 26-786. Covered devices are those weighing over 4.4 lbs (about 2kg) with connectivity of 200kbps or higher; new models will no longer be able to obtain the equipment certification required for import and sale in the US.
Existing certifications aren't retroactively revoked. Already-certified machines can continue to be sold and used, but modifying the hardware would require an individual waiver. This isn't an outright ban — it's a measure that closes the entry point for new devices.
Where that line actually falls becomes clear looking at an NVIDIA announcement. At GTC Taipei on June 1, 2026, the company unveiled its Isaac GR00T reference humanoid robot. Its body is Unitree's H2 Plus chassis (about 183cm, 68kg, 31 degrees of freedom), fitted with a Sharpa Wave five-finger hand (22 degrees of freedom) and Jetson AGX Thor T5000, for 75 degrees of freedom overall. It will be supplied by Unitree starting in late 2026, with adopters including Ai2 and Stanford Robotics Center among four institutions listed. The body that American researchers are set to use as their standard reference is built on the chassis of a manufacturer the US itself added to its import restrictions in July.
And this is where the real question about the pace of price declines comes back into focus. Learning curves fall based on how many units of the same design are built in the same factory. If supply chains split in two along regulatory lines, each side's cumulative volume loses a significant chunk of its denominator, delaying the arrival of the $240 joint. Even if the Bank of America Institute's 14% annual rate holds, the $35,000 figure won't reach the $15,000-range median until around 2031 — and it would take even longer from there for that to show up in actual retail prices.
So the numbers worth tracking next aren't Goldman's revision magnitude or its TAM figure. They're how far Counterpoint's application breakdown moves intelligent manufacturing's 13% and warehouse logistics' 5% upward, where Unitree's average selling price bottoms out, and how much Harmonic Drive Systems' humanoid-bound shipments accumulate in reducer unit counts. Once a single joint reaches $240, humanoid robots stop being capital equipment that needs a formal approval process for every unit and start becoming a tool you simply add more of as a process demands. But as Melonee Wise pointed out in an interview with IEEE Spectrum, nobody has yet found an application that requires thousands of units at a single site. Whether demand is actually waiting once prices come down remains a separate question from the cost math itself.