← Back to Research
Humanoid Robotics

Humanoid Robots 2026: 100,000-Unit Guidance, but Valuations Have Run Ahead of Output

FutureX Research · AI Lab · 2026.06.23 · 24 pp · preview 6 pp · Revised 2026.11.02

🎧

Listen · Audio Summary

5-8 min · AI narration in English · abstract + all key findings

Bottom Line

2026 is the hardware year for humanoids: capacity is rising, data and revenue lag, and valuations moved first.

Why now
By the October 9 close, Unitree sat about 60.7% below its August 19 debut-day open: public markets have started repricing humanoids on delivery.
What it means
For founders and investors: anchor valuations to the repriced public level, and expect moats to move from hardware specs to data assets.

Abstract

2026 is the hardware year for humanoid robots: capacity is rising, data and revenue lag, and valuations moved first. MIIT said in early July that China's full-year output could exceed 100,000 units, yet on July 22 Tesla said Optimus had not formally started production. Unitree closed its August 19 STAR Market debut up 460.34%, and by the October 9 close it sat about 60.7% below its debut-day open. SoftBank is in talks to take control of 1X at about $6 billion, below the roughly $10 billion valuation reported last year. Only about 500,000 hours of high-quality physical interaction data exist globally, a gap above 99%. Compiled from public information; not investment advice. (Data current as of 2026-10-11.)

Key Points

  • 01Unitree listed on Shanghai's STAR Market on August 19 and closed day one up 460.34%; on October 9 it closed at RMB 432.00, about 60.7% below its RMB 1,100 debut-day open (Public market data).
  • 02MIIT said in early July that China's 2026 humanoid output could exceed 100,000 units; China shipped about 14,400 in 2025, and Morgan Stanley expects 50,000 shipments in 2026 (Reported).
  • 03Tesla has ordered components for about 15,000 Optimus units in 2026, The Motley Fool reported September 14, less than a third of the low end of its official 50,000-100,000 target (Reported).
  • 04SoftBank is in talks to buy a majority stake in 1X Technologies at about $6 billion, below the roughly $10 billion valuation 1X reportedly sought last year (The Information, August 26; Reported).
  • 05Only about 500,000 hours of high-quality physical interaction data existed globally entering 2026, while a general embodied model needs tens of millions, a gap above 99% (Pedaily, August 12; Reported).

I. Production (1): MIIT Guides 100,000 Units; Banks Forecast 30,000 to 50,000

MIIT officials said at the WAIC press conference in early July that China's humanoid robot output could exceed 100,000 units in 2026 (Reported). China shipped about 14,400 units in 2025, and the world about 18,000 to 20,000. Reaching 100,000 would mean the industry moves from building robots to mass-producing them.

Banks are more cautious. The gap between guidance and forecasts is the key variable to check in the second half:

Source2026 China outlookEvidence
MIITOutput could exceed 100,000 unitsReported
GGII (Gaogong Robotics Industry Research Institute)62,500 unitsReported
Morgan Stanley50,000 units shipped (raised from 28,000)Reported
UBSAbout 30,000 unitsReported
Baseline: 2025 actualAbout 14,400 units shipped in ChinaReported

Exports are growing. Q1 robot exports reached RMB 11.32B, and humanoid exports rose 210% year on year, reaching 148 countries and regions (Reported).

The event calendar is dense. WAIC in Shanghai, which opened July 17, drew 1,100+ companies, 3,000+ exhibits and 300+ global debuts, with production models and live automotive and logistics lines at the center. The World Robot Conference in Beijing Yizhuang (Aug 19-23) previewed 150+ more debuts and a global application-exploration program that places mass-produced robots in real settings for validation (Reported).

I. Production (2): AgiBot's 15,000th Unit Puts China's Capacity Race in Five Figures

China's leading makers have moved from thousands of units to tens of thousands. AgiBot's 15,000th general-purpose embodied robot rolled off the line in late June; three years ago its annual output was six units (Reported). The company says its self-built "A-Chain" standardized supply chain supports order-driven flexible capacity above 100,000 units a year, and it targets shipments in the tens of thousands for 2026 (Per company disclosure).

Field validation is moving in parallel. In June, eight Genie G2 robots completed a six-day livestreamed run on Longcheer's production line, handling high-precision inspection such as multimedia-interface and audio testing on their own at a 99.99% success rate (Reported).

Prices are falling. Unitree shipped 5,500+ humanoids in 2025, ranking first globally (Reported). At WAIC it showed the H2 (1.82m, 70kg, 3-hour runtime), launched in October 2025, and the GD01 rideable transforming mech, launched in May; the dual-arm R1 series, introduced in April, now sells for under RMB 30,000 (Per company disclosure). Leju posted 2025 revenue of RMB 258M with a three-year CAGR of 118.68%, and is pursuing a ChiNext listing on a price-for-volume strategy (Reported).

Lower prices and higher capacity are turning humanoids from demos into industrial goods.

I. Production (3): In Late July Tesla Said Optimus Had Not Formally Started Production; Figure Has Delivered 350+ Figure 03s

The two overseas leaders' actual ramps say more than their marketing.

Tesla took 46 days to strip out Fremont's Model S/X line and install the first-generation Optimus line, and Musk was on site July 1 to confirm it was running (Reported). But the July 22 Q2 call showed revenue up 26% year on year and operating profit down 57%, and said Optimus had not formally started production: the first-generation line was still being installed, with production expected later this year (Per company disclosure). Musk said Optimus has about 10,000 all-new parts and needs a supply chain built from scratch, calling it "the hardest product to scale manufacturing" in Tesla's history.

Figure's numbers are more concrete. It has delivered 350+ Figure 03 units, and its line has gone from one robot a day to one an hour (Reported). Since late June, Figure 03 has handled logistics sequencing at BMW's Spartanburg plant, where Figure 02 earlier worked on body-shop tasks for 30,000 X3s (Reported).

MetricTesla OptimusFigure 03
Status, late JulyNot yet in formal production; first-gen line still being installed350+ units delivered
2026 target / capacityOfficial target of 50,000-100,000 unitsBotQ phase one rated at 12,000 units a year; 100,000 over four years
UseFirst units for training-data collection, not customer deliveryLogistics sequencing at BMW Spartanburg since late June

The entry barrier keeps falling. NVIDIA's Isaac GR00T reference humanoid, built on Jetson Thor and Unitree's H2 Plus and unveiled in June, is expected to go on sale through Unitree by year-end (Per company disclosure).

Outside Views: Jim Fan and LeCun Both Write Off VLA and Split on What Replaces It

Jim Fan (Director of Robotics Research, NVIDIA; Sequoia AI Ascent, May; Verified): he asked the room for "a moment of silence for teleoperation." A robot is capped at 24 hours a day, about 3 in practice, so teleoperation cannot scale; it is already under 0.1% of training data. His replacement is egocentric human video ("the new FSD data," scalable to 10 million hours a year) plus world action models, with a physical Turing test passed within 2–3 years.

Yann LeCun (Turing Award laureate; May; Reported): the VLA (vision-language-action) route is "largely considered a failure." Agentic LLMs in the physical world are "intrinsically unsafe" because autoregressive models cannot anticipate the real consequences of actions. He expects the field to concede a paradigm shift by early 2027.

The two rarely agree, but both treat VLA as dead; they split only on what replaces it: world action models or hierarchical world models. Set that beside this report's capital data. The architecture has not converged, yet SoftBank is negotiating for control of 1X at $6 billion, and Unitree's market value reportedly fell about 30% in two weeks. Valuations moving before the technology settles is the risk structure this report keeps flagging.

FutureX Position · Open Source Breaks the Deadlock: 2026 Is the Hardware Year for Humanoids, and 80 Points Is Zero (Xiamen keynote, 2026-09-03)

The Xiamen keynote reinforces this report's conclusion and adds a mechanism. The report argues that technical routes have not converged while valuations ran ahead. FutureX puts it more bluntly: 2026 is the hardware year for humanoids, and the intelligence year has not arrived. Its rule is "80 points equals zero": a robot that finishes eight tenths of a task has finished nothing.

The mechanism is verifiability. In FutureX's reading, the order in which AI changes an industry equals verifiability times feedback cycle. Code and content are verified automatically, with feedback in seconds, so they moved first. Medicine and manufacturing keep a human in the loop, with feedback in months, so they move last. Robotics sits late in that order; FutureX dates the robotics-and-logistics window to 2025–26. Slow verification means slow data, and slow data means no route can prove itself yet. SoftBank's talks to take control of 1X at a lower valuation, recorded in this report, fit that formula.

Shipments are climbing. Humanoid shipments reached 18,000 units globally in 2025, and GGII (Gaogong Robotics Industry Research Institute) projects 62,500 in China for 2026 (Reported). FutureX's focus sits away from the body, on dexterous hands and data, with no part in the full-body arms race.

The keynote also ran a hardware tally:

Source of gainCompute efficiency gain, 2012–2023
TotalAbout 1,000x
Number formats (quantization)16x
Sparsity2x
Instruction sets12.5x
Process node (28nm to 5nm)2.5x

FutureX's read is that the next 1,000x will come from architecture: silicon photonics interconnect, compute-in-memory and 3D integration. The human brain runs on 20 watts; a robot cannot carry a 1,000-watt GPU.

Full deck: /reports/open-source-breakthrough (first 5 pages public).

Latest developments, newest first (data current as of 2026-10-11)

  • 2026-10-09 · Wu Hong, a senior inspector at the Guangdong Department of Industry and Information Technology, said during the provincial advanced manufacturing conference that Guangdong aims to produce more than 80,000 humanoid robots a year by 2028, equal to 80% of the 100,000 units MIIT expects nationally for 2026; the province expects about one third of China's 2026 humanoid output, and more than 3,000 humanoids could be working in Guangdong factories this year · Verified (China News Service; also in Caixin's October 10 morning briefing)
  • 2026-10-09 · Unitree (688836) hit an intraday low of RMB 419.21, its lowest since listing, then closed at RMB 432.00, up 0.98%, for a market value of about RMB 174.7 billion; that is still about 186% above the RMB 150.80 offer price and about 60.7% below the RMB 1,100 debut-day open · Public market data (Tencent Securities)
  • 2026-10-09 · Daily humanoid rental rates fell from the RMB 10,000 level in early 2025 to RMB 2,000-3,000 in early 2026, and are now below RMB 1,000; on Xianyu and similar platforms, one merchant lists a Unitree humanoid from RMB 289 a day. Rentals had been hard to book after the 2025 Spring Festival Gala · Reported (STAR Market Daily, via IT Home)
  • 2026-10-08 · On the first trading day after the National Day break, Unitree closed at RMB 427.80, down 5.02% from RMB 450.40 on September 30 · Public market data (Tencent Securities)
  • 2026-09-30 · Unitree closed at RMB 450.40 with an intraday low of RMB 446.50, both the lowest since listing at the time; the close was about 199% above the offer price, about 59.1% below the debut-day open and about 12.5% below the RMB 514.98 close on September 18 · Public market data (Eastmoney)
  • 2026-09-30 · Figure released a video of retired F.02 robots lowered into a 75-ton electric arc furnace in Imatra, Finland. Its blog said maintaining F.02 made less sense as the F.03 fleet grew, and manual disassembly would take engineering time needed for F.04; F.03 units are deployed with BMW Manufacturing and retail group Catalyst Brands · Reported (Humanoids Daily)
  • 2026-09-28 · Unitree fell 5.81% to close at RMB 459.65 · Public market data (Eastmoney)
  • 2026-09-28 · Tesla has lifted Optimus output about tenfold in recent months, from a few dozen units a week in Q2 to several hundred a week in August, and aims for more than 1,000 a week by end-2026. The bottleneck is the hand and forearm: each has 100+ screws and small parts that workers assemble by hand, and imprecise alignment at several stations means more robots need rework after leaving the line. Tesla still relies on Chinese suppliers for parts and plans to lease, rather than sell, its first robots to commercial customers · Reported (The Information, via The Next Web)
  • 2026-09-18 · Unitree closed at RMB 514.98 for a market value of RMB 208.289 billion, about 241% above the RMB 150.80 offer price and more than 53% below the RMB 1,100 intraday high of August 19; Sina Finance's headline called it a halving · Public market data (Sina Finance)
  • 2026-09-17 · Unitree rose more than 7% intraday, and its market value returned to RMB 200 billion · Public market data (Sina Finance)
  • 2026-09-14 · Citing early-September supply-chain reports, Tesla has ordered enough components to build about 15,000 Optimus units in 2026. Musk said at Davos in January that Tesla could ship its first commercial Optimus robots to external customers by 2027 and that 2027 output could reach one million; most analysts expect 100,000 to 500,000 units in 2027, mainly for Tesla's own factories and pilots. Musk has repeatedly put the at-scale price at $20,000 to $30,000 · Reported (The Motley Fool)
  • 2026-09-04 · Index, Figure's data-collection app, reached 69,943 weekly active users in the week ending September 4 and takes in about 35 minutes of human task video per second worldwide; at its late-August public launch it had about 44,000 participants and 30 minutes per second, so weekly actives rose about 59% and the upload rate about 17% in two weeks · Reported (Humanoids Daily, September 8, citing Figure founder Brett Adcock)
  • 2026-09-04 · Agility Robotics filed a Form S-4 with the SEC for its SPAC merger with Churchill Capital Corp XI: 2025 net sales of $1,781,967 and a net loss of $138,086,332, about 77 times revenue; its $300 million multi-year backlog comes from one related-party customer, which receives 453 common stock purchase warrants per robot ordered · Reported (Humanoids Daily, September 7)
  • 2026-09-04 · Unitree closed the week at RMB 530.3, down 9.35% from RMB 585.0 a week earlier, after an intraday high of RMB 591.0 on September 1 and a low of RMB 528.85 on September 4; its market value at week's end was RMB 214.487 billion · Public market data (Phoenix Finance, September 6)
  • 2026-08-31 · By the end of August, Unitree's market value was reported to have fallen about 30% from its first-day close · Reported
  • 2026-08-26 · SoftBank is in talks to acquire a majority stake in OpenAI-backed humanoid maker 1X Technologies at a valuation of about $6 billion; terms may still change. That is below the roughly $10 billion valuation 1X was reported to be seeking during a $1 billion raise last year, a retracement of about 40%. 1X's flagship is NEO, a home humanoid announced in October 2025 at an early-access price of $20,000, with a planned $499-a-month subscription · Reported (The Information exclusive; syndicated by Reuters and others)
  • 2026-08-19 · Unitree (688836.SH) listed on the STAR Market: 40,446,434 shares at RMB 150.80, an issue market value of about RMB 60.99 billion. It opened at RMB 1,100, up 629.44%, worth about RMB 444.9 billion, and closed at RMB 845, up 460.34%, worth about RMB 341.8 billion, 5.6 times the issue value; a winning 500-share lot showed a paper gain of about RMB 347,000 at the close · Verified (Shanghai Stock Exchange listing announcement and market data)
  • 2026-08-19 · 37 allocated insurance institutions may hold more than RMB 4.7 billion in paper gains · Reported (Yicai)
  • 2026-08-12 · Chinese embodied-AI startups raised RMB 93.5 billion in the first half of 2026, about five times the year-earlier figure; the working consensus holds that a general embodied model needs tens of millions of hours of high-quality real interaction data, against a global pool of only about 500,000 usable hours entering 2026, a gap above 99% · Reported (Pedaily)

What This Means for Founders

Only deliveries and renewal rates can move a valuation from narrative to cash flow, such as renewals on 1X's $499-a-month NEO subscription. Embodied data has to be captured frame by frame in the physical world, so defensibility is shifting from hardware specs to data assets and capture systems; Figure's Index grew weekly actives about 59% in two weeks. STAR Market Daily reports that daily humanoid rental rates have fallen below RMB 1,000, so demo and novelty demand no longer supports pricing; revenue has to come from factory and other production deployments. Independent funding is tightening: 1X is in talks to cede control at about $6 billion, and strategic buyers are becoming the main clearing path.

What This Means for Investors

Money is plentiful, yet independent funding is tightening: China's embodied-AI startups raised RMB 93.5 billion in the first half of 2026, about five times a year earlier, while 1X is in talks to cede control at about $6 billion. Anchor valuations to the repriced public level: on October 9 Unitree sat about 60.7% below its debut-day open, so day-one peaks no longer apply. Deliveries have not yet turned into revenue at the robot makers, with Agility's 2025 loss about 77 times revenue, while data infrastructure is becoming an investable layer of its own. Signals to watch: Unitree's 2026 third-quarter revenue growth, whether Tesla reaches 1,000 Optimus units a week by year-end, and whether a second leading company accepts a down-round control deal. This is not investment advice.

Where We Could Be Wrong

  • 01If Tesla reaches 1,000 Optimus units a week by end-2026 (about 50,000 a year annualized) and hands no longer need manual assembly, our view that output is bound by dexterous hands and line precision does not hold.
  • 02If no second leading humanoid company accepts a 1X-style down-round control deal by end-2026, the tightening of independent funding is an isolated case.
  • 03If daily rental rates recover above RMB 1,000 and Guangdong meets its target of 3,000 factory deployments this year, our view that demand cannot support pricing and the installed base is underused needs revising.

Key Questions

When is Unitree's IPO subscription, and at what valuation?

Unitree held book-building on Aug 5, 2026 and subscription on Aug 10, issuing 40.45 million shares (10% post-issue) at RMB 150.80 to raise about RMB 6.1 billion, a post-issue value of roughly RMB 61 billion; it listed Aug 19. Registration took effect July 2 — just 104 days from acceptance, a STAR Market record — and institutions see potential for a RMB 100 billion market cap after listing.

How many humanoid robots will be produced in 2026, and where does Tesla Optimus stand?

In early July MIIT projected China's 2026 humanoid output could exceed 100,000 units (vs ~14,400 China shipments in 2025), and Morgan Stanley raised its 2026 China forecast from 28,000 to 50,000 units. Tesla's July 22 Q2 call confirmed zero Optimus output to date with the Fremont line still being installed, against a 2026 target of 50,000-100,000 units; Figure has delivered 350+ Figure 03 units, now working at BMW Spartanburg.

Is Kimi K3 open-sourced, and what is Moonshot AI's latest valuation?

Yes. Moonshot released the 2.8-trillion-parameter Kimi K3 on July 16 (KDA hybrid linear attention, 1M-token context) and opened full weights on July 26 — a day early — under a modified MIT license. It then reportedly closed a $3.5 billion round at a $35 billion post-money valuation, with a pre-IPO round planned for August at a rumored valuation of up to $50 billion and a Hong Kong Chapter 18C listing in view.

Watch & Listen

In China: search WeChat Channels for 「倩姐投AI」; full library → Qian on AI

Sourcing and standards

Compiled from public sources; data current as of 2026.10.11. The text separates four evidence grades: verified, reported, per company disclosure and public market data, and states the derivation behind every estimate. When we get something wrong, the correction is written into the report body with the original call left visible, and logged publicly.

Research standards & corrections →

📄 Full Report

Full report: 24 pages · provided to professional investors & partners only

This is the public preview. The full report includes the sections below. For compliance reasons it isn't posted publicly or offered as a free download. To request a copy, contact the FutureX team.

  • 🔒II. Capital Markets: The IPO Window Opens Wide — Unitree's Countdown and the Valuation Rift
  • 🔒III. The "Brain" Layer: Kimi K3 Weights Land Early and Brain Costs Collapse
  • 🔒IV. AI Hardware Wars: Apple v. OpenAI and Spillover Signals from the Giants
  • 🔒V. Compute and Energy: 3.2GW Power Deals and Mobile Turbines — the Hidden Upstream Bottleneck
  • 🔒VI. Value Chain Map: Who Benefits, in a Neutral Framework
  • 🔒VII. Risk Factors: Zero Output, Price Wars and Diverging Forecasts
Request the full report →

Where we stand on this

Questions people ask next

Building in this space, or want to discuss this report? Write to us. We usually reply within 48 hours →

Related Research

Industry research from FutureX Capital's AI Lab, compiled from public information; not investment advice; contains no fund performance, AUM, or offer to raise capital.