For most of the last decade, asking how many humanoid robots a company would build next year was a polite way of asking nothing at all. The answer was always "a few" — a handful of prototypes, a couple of pilots, maybe a small batch for research labs. 2026 is the first year that question stops being polite and starts being operational. Several manufacturers are actually committing to production numbers. And 2027 is when those numbers, if they hold, start to look like an industry rather than a science project.
What follows is a side-by-side comparison of what the major players have publicly announced, what their factory capacity actually allows, and how realistic each number looks against past delivery records. The estimates below are expected production — units rolling off the line in the calendar year — not cumulative fleet, not orders, not vague "by 2030" promises. Where companies have given a range, we've kept the range. Where they've given a single ambitious figure, we've paired it with what the field is willing to bet they'll actually hit.
The Numbers, Side by Side
| Manufacturer | Robot | 2026 (expected) | 2027 (expected) | Approach |
|---|---|---|---|---|
| Tesla | Optimus | 5,000–12,000 | 50,000–100,000 | In-house, factory-first |
| Xpeng | Iron | 2,000–5,000 | 20,000–30,000 | In-house, EV transfer |
| Figure AI (BMW) | Figure 03 | 8,000–12,000 | 25,000–40,000 | BotQ plant, automaker pilots |
| Boston Dynamics (Hyundai) | Atlas (electric) | 200–500 | 1,500–3,000 | Hyundai plants, in-house |
| Unitree | G1 / H1 / R1 | 8,000–15,000 | 25,000–50,000 | Multi-SKU, low price |
| Agility Robotics | Digit | 1,500–3,000 | 8,000–10,000 | RoboFab, logistics-first |
| Apptronik (Mercedes) | Apollo | 1,000–2,000 | 5,000–10,000 | Jabil partnership, enterprise |
| 1X Technologies | Neo | 1,000–3,000 | 10,000–20,000 | Consumer-first launch |
| UBTech | Walker S2 | 1,000–2,500 | 5,000–10,000 | Chinese auto OEMs |
| Fourier Intelligence | GR-3 | 500–1,500 | 3,000–6,000 | Research & rehab focus |
| Aggregate (range) | ~28,000–56,000 | ~150,000–280,000 | ||
Figures consolidate public announcements, partner contracts and reported factory capacity as of Q2 2026. Ranges reflect both stated targets and the realistic delta between announcement and delivery. Numbers will move as the year develops.
The combined upper end of these announcements puts the industry on a path to a quarter of a million humanoids built in 2027 alone. Even at the lower end, it's a ten-fold expansion over 2025. The decade's defining number is starting to take shape.
Tesla — Optimus
Musk has been the loudest voice in the room and also the one whose numbers move the most. The original "1 million units a year" target slid into "thousands in 2025, tens of thousands in 2026" and is now realistically in the 5,000–12,000 band for 2026, with the V3 design ramping at Fremont and Giga Texas. The 2027 jump is the one to watch: Tesla has staked a public claim that Optimus is the most important product in the company's history, and the only way to back that up is to start manufacturing at a scale no one else can touch. If they hit even the lower end of 50,000 units, they redefine the cost curve for the entire industry.
Xpeng — Iron
Iron is the surprise of the field and the clearest test of the EV-to-robot transfer thesis. Xpeng has confirmed mass production of the next-gen Iron starting late 2026 from its Guangzhou facility, with the first units sold to enterprise clients before any consumer launch. The 2,000–5,000 figure for 2026 is essentially a soft launch; the real test is whether they can hit the 20,000–30,000 range in 2027 using the same supply chain that produces their vehicles. If they do, Xpeng becomes the first non-Tesla manufacturer to break the five-figure annual barrier — and the first to do so at a price point (around $27,000) that makes the math actually work.
Figure AI (BMW) — Figure 03
Figure's BotQ plant in San Jose was designed for 12,000 units a year of capacity from day one, and the four-year, 100,000-unit framework agreement with BMW gives them the demand certainty most of the field lacks. The Figure 03 is already operating at BMW Spartanburg, and the 8,000–12,000 estimate for 2026 essentially saturates first-year BotQ throughput. The 2027 jump assumes a second production line and additional automotive partners beyond BMW. Figure has been the most disciplined of the new entrants about not over-promising — which makes their numbers among the most credible in the table.
Boston Dynamics (Hyundai) — Atlas
Atlas is the brand everyone knows and the production line nobody runs. Boston Dynamics has been deliberate to the point of frustrating its parent: the electric Atlas only entered limited Hyundai plant pilots in late 2025, and the company has explicitly chosen capability depth over volume. 200–500 units in 2026 is a Hyundai-internal number, not a commercial one. The 2027 figure assumes a controlled expansion to additional Hyundai/Kia facilities, but Boston Dynamics is not playing the same game as Tesla or Figure. They're building the most capable platform and letting volume follow when the engineering allows. That bet may still pay off — but not in the units column for the next two years.
Unitree — G1, H1, R1
Unitree is the volume story most Western coverage misses. They already ship G1 and H1 at meaningful numbers to research labs, universities and developer kits, and the upcoming R1 pushes the price floor below $6,000. Their Hangzhou production has the capacity to absorb a 3-4× expansion in 2026 without capital expansion, and the multi-SKU strategy spreads risk across price points that no single Western competitor covers. The 8,000–15,000 range for 2026 looks conservative if developer demand for G1 holds. The 2027 figure assumes Unitree starts to compete for enterprise and light-industrial deployments, which is where the H1 successor is positioned.
Agility Robotics — Digit
Agility's RoboFab in Salem, Oregon, was the first dedicated humanoid factory in the world when it opened, with stated capacity of 10,000 units a year at full build-out. The 1,500–3,000 figure for 2026 reflects current ramp, with Digit deployed across GXO Logistics and Amazon facilities. Agility's logistics-first strategy is a different bet from the rest of the field — they're not chasing the consumer market or the assembly-line market, they're chasing warehouses, where ROI is fastest and tasks are most uniform. If that bet is right, the 2027 figure could be conservative.
Apptronik (Mercedes) — Apollo
Apollo's production is being scaled in partnership with Jabil, the contract manufacturer that builds for Apple and others — a structural advantage that lets Apptronik ramp faster than a pure in-house factory would allow. The Mercedes pilots have been live long enough to generate operational data, and the 1,000–2,000 estimate for 2026 reflects the first commercial deliveries beyond the Mercedes-Benz Alabama plant. The team's NASA heritage and manipulation stack quality are well-regarded; the 2027 figure assumes they convert that into multiple enterprise contracts beyond the automotive sector.
1X Technologies — Neo
1X is the consumer outlier. While everyone else is racing for factories and warehouses, 1X has launched Neo as a home robot — soft-clad, quiet, designed for living rooms rather than loading bays. The 1,000–3,000 estimate for 2026 is essentially a beta-program scale; the 10,000–20,000 jump in 2027 depends on whether the consumer launch generates the kind of waitlist demand that OpenAI-backed marketing suggests. 1X has the most uncertain numbers in the table — but also the highest variance: if the consumer thesis is right, they could easily exceed the upper end.
UBTech — Walker S2
UBTech has pivoted aggressively from companion robots to industrial humanoids, and the Walker S2 has been deployed at NIO, BYD and Geely facilities for multi-shift assembly testing. The 1,000–2,500 figure for 2026 reflects the Chinese government's explicit policy support for domestic humanoid adoption — a tailwind no Western manufacturer has. The 2027 figure assumes the early pilots convert into volume orders, which is plausible given how integrated UBTech has become with the Chinese auto sector. Less mediatic in the West, more operational than most assume.
Fourier Intelligence — GR-3
Fourier comes from a rehabilitation-robotics background and has carried that precision into their humanoid line. The GR-3 is positioned for research, medical and light-service applications rather than the heavy-duty factory deployments that dominate the rest of the field. The 500–1,500 estimate for 2026 reflects a deliberate choice to stay in that niche rather than compete head-on with Unitree on price or Tesla on volume. That focus is a feature, not a limitation — but it caps the production numbers accordingly.
The Honest Caveat
Every number in the table above carries the same asterisk: the gap between humanoid announcements and humanoid deliveries has historically been large. Tesla's original 2025 target was an order of magnitude above what they actually shipped. Figure's early roadmaps were more aggressive than what BotQ has so far produced. The right way to read this comparison is not as a forecast but as a map of intent — what each manufacturer is publicly committing to, and what the structural constraints of their factories actually allow. The companies that hit the upper end of their ranges in 2026 will be the ones to bet on for 2027. The rest will quietly revise their projections, the way Musk has done every year for the last four. Either way, the numbers are real enough now to compare. That alone is the news.