The robots coming next — the ones that will define whether this decade delivers on five decades of science fiction promise. Here we track announcements, specs, and our take on every major upcoming launch.
BYD confirmed its first humanoid for August 2026 at its Di Space centre in Zhengzhou — aimed at dealership floors, two to three per store. What BYD has actually confirmed, the unofficial 1.61 m / 58.5 kg / 31-DOF spec sheet, the three claims BYD denied outright, and why nobody outside China has demonstrably seen it yet.
Unitree's strategic placement drew DeepSeek (¥140m+, three-year lock-up) and Tencent, and its retail lottery was the tightest in STAR Market history. But AgiBot outshipped Unitree in 2025 and has no direct listing. The four robot makers you can actually buy, AgiBot's ¥2.1bn side door into Shanghai, the 20–30 company IPO pipeline, and why the best manufacturers are the hardest to own.
Unitree priced its Shanghai STAR Market IPO at ¥150.80 a share on 6 August 2026 — ticker 688836, a ~¥61bn ($9.0bn) valuation, a 10% float and an 8,289× oversubscribed retail book. The full numbers, the listing timetable, why foreign retail investors cannot buy it directly, the ETF and Hong Kong routes that do exist, and my case that $9 billion is cheap for the industry's leading manufacturer.
Beijing now requires 1,400+ schools to teach at least eight class hours of AI to every grade, and robots are the hardware doing it — block programming, iFlytek image-recognition units, a robotic arm driven by a smart headband. The mandates city by city, the research that says robots only help when a teacher is present, and what it means if you're buying for a child.
Humanoid design is splitting in two: engineering-first machines like the Unitree G1 (1.27 m, 35 kg) that move best, and human-proportioned, soft-covered robots like Tesla's Optimus and XPeng's IRON (1.78 m, 82 DOF, bionic spine) that look best. The three design schools, a league table of 12 robots, and what the median humanoid looks like by 2028.
The World Robot Conference 2026 runs August 19–23 in Beijing with 300+ exhibitors, 2,000+ exhibits, 150+ product debuts and a first-ever 49-company state-owned-enterprise pavilion. Dates, venue, the World Robot Contest, what's new, and what to watch.
On July 28, 2026 the FCC added foreign-made humanoids and quadrupeds to its Covered List, blocking new models from being imported, marketed or sold in the US. It doesn't touch what you already own — but it slams the door on new Chinese-made robots. What the rule does, and what it means for buyers.
China's Beijing Humanoid Robot Innovation Center (X-Humanoid) built Tiangong — the marathon-winning humanoid it open-sources to the whole industry, from motion-control code to the RoboMIND dataset and the WoW world model — while drafting the national standards. How the state-backed open-source playbook works.
China makes the reducers, actuators and rare-earth magnets inside nearly every humanoid — even Tesla's Optimus. Morgan Stanley puts China at ~63% of the humanoid supply chain, and stripping Chinese parts out of Optimus could nearly triple its cost. Who makes what, and the one bottleneck China hasn't cracked.
Musk called Optimus the hardest thing Tesla has ever built and warned of a "flat and long" ramp — and Tesla stock fell ~14%, while Unitree heads for a record IPO. Why the difficulties are the whole industry's, and the honest timeline: not 50 years, more like 5–10 (I'd bet 2–5).
AgiBot's flagship from its four-robot WAIC 2026 launch: a 1.74 m service humanoid with 51 degrees of freedom, up to 5 kg per arm, up to 8 hours of runtime and NVIDIA Thor compute. Full specs and an honest verdict.
Tesla promises a million Optimus robots a year — but as of July 2026 it hadn't started. The companies actually shipping humanoids are Unitree, Figure and Agility. The honest, evidence-ranked production scoreboard.
At its Munich EV launch, XPeng put fresh numbers on its IRON humanoid: more than 1,000 units a month by the end of 2026, mass production at year-end, and a 2027 global rollout — starting with robots that answer questions in its own showrooms and offices.
The first humanoid robot that swaps its own battery — in about three minutes — to work 24/7. Specs, the ~$160,000 price, its BYD and Foxconn deployments, and an honest verdict.
A bipedal, general-purpose humanoid you can't buy yet versus a wheeled, rope-driven manipulation specialist on sale now for $96k–$150k. Full specs side by side, the teleoperation question over both, price, and an honest verdict on which is which.
The Tesla AI5 chip is the custom inference brain behind Optimus Gen 3. Musk claims ~8x the compute of AI4 and roughly one Nvidia H100 per chip — but it only taped out in April 2026 and won't reach volume until 2027. AI5 specs, the AI4 and H100 comparison, and what it means for the robot.
The rifle-carrying robot dogs going viral from China are real, but teleoperated — a human still pulls the trigger. That is precisely why now is the moment to bar robots from carrying or using any weapon, enforced both in their software and in binding international law, before the machines can decide for themselves.
A thought is electricity and chemistry — physical, and therefore measurable. Neuralink already reads intent through the skull; AI already decodes the gist of imagined speech without one. My on-the-record prediction: before 2100, robots will read human minds, and we'll talk to them without words — Asimov's R. Daneel, made real.
We forecast 28,000–56,000 humanoids built worldwide in 2026. Six weeks later Morgan Stanley raised its 2026 number again — to 50,000 for China alone, its second hike this year. A side-by-side comparison: we're more conservative on near-term China, far bolder on 2035, and the bank's number has only ever moved up.
Dot-com, mobile, crypto, AI — every wave had a moment to buy in early. Robots are next, and I think 2026 is robotics' Bitcoin-in-2010 moment. My on-the-record forecast: a ~$1.5 trillion humanoid market by 2035, benchmarked honestly against Goldman, Morgan Stanley, Citi and BofA.
Optimus Gen 3 is the production-intent humanoid built around Tesla's new AI5 chip and running Grok — but the chip meant to be its brain only taped out in April 2026 and won't reach volume until 2027. The reported specs, the 22-DOF tendon hands, the $20k–$30k target price, and why the body is running ahead of the brain.
No single humanoid is production-ready at everything. So we took the best subsystem from every robot in our database — Atlas's 50 kg arms, Tesla's 22-DOF hands, Figure's Helix brain, the DR02's all-weather body, Digit's warehouse track record — and assembled the perfect production humanoid, part by part. Then we explain why it can't be built.
None of the most hyped humanoids of 2026 are on sale to the public yet. Tesla Optimus Gen 3, Figure 03, 1X NEO, Unitree H2 Plus, Honor A1 and five more — the ten you can't buy, compared on specs, price and availability.
Unitree goes one step further again — this time allying with NVIDIA. From the $4,900 R1 to the Isaac GR00T-powered H2 Plus, the full Unitree humanoid lineup compared head-to-head on specs, price and purpose.
Pending final MIIT approval, China has begun assigning every humanoid a unique 29-character digital ID — a lifecycle passport tracked from factory to scrapyard. What the number means, why the world should align on one standard like USB-C, and what is still missing.
The railroad. The automobile. The dot-com. AI. Every generation gets one trend it fails to see coming. The next one is robots — and the math behind it (robots per human, $10,000 a unit, billions of units, the next Nvidia) is staggering once you actually run it.
Boston Dynamics Atlas has the most capable humanoid ever built: 56 DOF, 50 kg payload, self-swapping battery. Tesla Optimus has the FSD stack, the manufacturing flywheel, and a price target that could redefine the market. Neither has a confirmed commercial price. Here's the full comparison.
The G1 has dropped. The Go2 keeps getting cheaper. The R1 is already under $5,000. An Altman Solon survey of 1,000+ consumers shows 69% won't pay more — and the math explains exactly why that ceiling won't move.
Honor A1 (Yuanqi Zai) danced at MWC 2026 and won the Best Gait Award at the Beijing half-marathon. Not the fastest — but possibly the most important humanoid for the consumer market.
The first humanoid robot to break the $5,000 barrier. Not full-size, not perfect — but the price point that starts the hardware-software flywheel and changes what every competitor now has to answer for.
Three robots. Eighty-one hours. One hundred thousand packages. Zero human interventions. What Figure AI streamed live this week is not another demo — it is a signal about the world we are entering, and most people did not fully understand what they were watching.
Two product categories shipped in seven days. Best-in-class motor skills. The only consumer-priced humanoid lineup. A live developer marketplace. Unitree just won the first battle of the humanoid war. The decade ahead decides which playbook they actually end up following — and the difference is almost entirely within their control.
Unitree just opened a marketplace where third-party developers can sell skills, moves and behaviours for its robots. The Fortnite emote parallel is the hook — the platform play underneath is the actual story, and it changes how a mid-cap robotics company competes with giants.
Unitree just unveiled a 2.7-metre, 500 kg transformable mecha you can climb inside — and yes, it looks exactly like the AMP suit from Avatar. Why a $650,000 piloted robot is impressive, why it does not fit the human world yet, and why the long-term bet is still real.
How many humanoid robots will Tesla, Xpeng, Figure, Boston Dynamics, Unitree and the rest of the field actually build in 2026 and 2027? A side-by-side look at announced production targets — and how realistic each one is.
The most interesting robots of the last decade did not come from billion-dollar labs. They came from crowdfunded indie teams. Here is the hall of fame — and why we put our own money on Stack-chan, the open-source desktop companion arriving on our desk in days.
Hardware will always belong to whoever can build it. Software, in the age of vibe coding, no longer needs to. The case for an open robot platform — and why the first manufacturer to make the move wins the decade.
It is weatherproof, sees in the dark with thermal sensors, and can call the police faster than a human dispatcher. And it wants to walk you home at night — just like the serenos of Seville did for three centuries.
A million units by end of 2025. A price tag that could eventually undercut the cost of a used car. Is this bold ambition or the closest the industry has ever come to delivering the robot Asimov dreamed of?
From electric vehicles to humanoid robots — Xpeng is transferring everything it learned about mass-market autonomy into Iron, a robot designed for factories first and homes next.