Who's Actually Mass-Producing Humanoid Robots in 2026?

The number everyone quotes about humanoid robots in 2026 is one million a year — Tesla's stated ambition for Optimus. The number that actually matters is smaller, quieter, and comes from companies you hear less about. Because here's the honest state of play: as of mid-July 2026, Tesla had not started mass-producing Optimus at all, while Unitree, Figure and Agility were already pushing real humanoids out of real factories. In this category, loudness and shipping are two very different things.

This is the production scoreboard as it really stands — who's building humanoids at volume today, who's deploying them into paying work, and who's still on the promise stage. It pairs with our production forecast versus Morgan Stanley's; here we're less interested in 2035 projections and more in what left an assembly line this month.

The headline number vs the number that ships

Tesla has turned a decade-old Model S line at Fremont into an Optimus assembly line and is raising a dedicated robot plant at Gigafactory Texas. The long-term goal — repeated by executives all year — is one million Optimus units annually, at a consumer-scale price of roughly $20,000–$30,000. It's a genuinely staggering ambition, and if any company on earth can bend a manufacturing curve, it's the one that already did it with cars.

But ambition isn't a run-rate. Elon Musk walked the newly installed Fremont line on July 1, 2026 and openly pushed back on suggestions that production had quietly begun. Tesla's own guidance pointed to a start in late July or August 2026, with the redesigned "V3" body — the one actually engineered for mass production — expected to be revealed over the summer. In other words, the robot that's supposed to ship a million a year didn't have its production design public yet at the point everyone was quoting the million-a-year figure. We went deep on that gap in our look at Optimus Gen 3 and why the brain lags the body.

"In humanoids, the press-release number is the ceiling. The number that leaves the loading dock is the one worth trusting."

Who's actually shipping right now

Strip out the projections and look at units that physically exist and do work, and the leaderboard reshuffles completely. Three names carry it.

Unitree — the volume king

Hangzhou's Unitree shipped more than 5,500 humanoid units in 2025, roughly a third of the entire planet's output, and carried that momentum into 2026 with new models. Its edge isn't a viral demo; it's price. A body like the R1 starts around $4,900, which is what makes volume possible in the first place. If you want the human story behind that, we profiled Wang Xingxing, the engineer who made Unitree's robots cheap, and compared the full Unitree lineup.

Figure — the Western factory bet

Figure is the clearest Western answer to the China volume story. The company says its purpose-built BotQ factory is producing the Figure 03 at roughly 55+ units a week, with the robot now moving onto BMW's line at Spartanburg after an eleven-month pilot with the previous Figure 02. Figure also carries a reported $39 billion valuation and announced a collaboration with OpenAI. Treat the weekly rate as a company figure rather than an audited one — but even discounted, it's the fastest Western scale-up in the field. Meet the man behind it in our profile of Brett Adcock, Figure's serial-founder CEO.

Agility Robotics — the deployment leader

If Unitree wins on units and Figure on factory pace, Agility's Digit wins on real-world mileage. Digit has moved more than 100,000 totes in live logistics work at GXO and signed a deployment deal with Toyota in Canada, and Agility announced plans to go public via a SPAC at a roughly $2.5 billion valuation in July 2026. Deployment longevity — robots that show up and work shift after shift — is a different, harder proof than a factory throughput claim, and it's the one Agility leads.

Rounding out the serious shippers: Boston Dynamics is delivering initial units of its electric Atlas to Hyundai while partnering with Google DeepMind on the brain, and China's UBTech is deploying its Walker S2 — the humanoid that swaps its own battery — at BYD and Foxconn. XPeng, meanwhile, has put a date on it: 1,000 IRON units a month by the end of 2026.

The scoreboard: loud vs shipping

Here's the honest read on the major names, sorted by how much real 2026 production evidence there is behind them — not by how often they trend.

None of this dismisses Tesla — a converted Fremont line and the FSD manufacturing flywheel are real advantages, and when Optimus does ramp it could ramp hard. It's simply a correction to the scoreboard: in July 2026, the robots actually being built at volume wear Unitree, Figure, Agility and UBTech badges, not a Tesla one.

Why China is the center of gravity

You can't read the 2026 production map without landing in China. Chinese firms accounted for roughly 85% of all humanoid robot shipments in 2025, and the pattern held into 2026: the cheapest bodies, the fastest iteration, and the densest supplier base for actuators and hands all sit in Hangzhou and Shenzhen. It's why we keep coming back to why robots are the next global trend after AI — and why the company that shipped more humanoids than anyone in 2025, AgiBot, is Chinese. The same week this scoreboard was taking shape, AgiBot used Shanghai's WAIC 2026 to unveil four new robots at once — a story we break down in our WAIC 2026 recap.

This is also where factory velocity matters more than any single unit count. A target tells you what a company hopes to build; the rate at which it can actually repeat the build tells you whether it will. We unpacked that distinction in how fast humanoids are actually being built, and it's the lens that separates a Unitree from a company still hand-assembling prototypes.

What this means if you're waiting to buy one

Here's the practical takeaway. "Mass-produced" and "buyable by you" are not the same thing in 2026. The humanoids genuinely rolling off lines in volume are mostly developer and industrial units — the Unitree R1 and G1, the AgiBot X2, the Booster T1 — the ones we rounded up in the humanoids you can actually buy in 2026. The mass-market consumer robot you've seen in the highlight reels — Optimus, Figure 03, 1X NEO — is a 2027-and-beyond story, no matter how the production headlines read today. The ten most hyped humanoids of 2026 still can't be bought by the public.

And there's a reason to be a little skeptical of every weekly-rate claim in this space: the same show-floor gap we documented at ICRA 2026, where "autonomous" robots were quietly driven by a human with a joystick, applies to production numbers too. A "55 units a week" figure from the company building the robots is a starting point for scrutiny, not a settled fact. The honest posture is to track deployments you can verify — totes moved, lines staffed, units in customers' hands — over targets you can't.

"Ask who's shipping, not who's talking. In 2026 the answer is a Chinese quadruped-maker turned humanoid giant, a Silicon Valley factory bet, and a logistics robot that just quietly moved its hundred-thousandth tote."

Frequently asked questions