Capital milestoneMos IPO / NetEase Finance · 17 Oct
Market reports indicate that AGIBOT has launched a new RMB 1–1.5 billion round at a post-money valuation of about RMB 50 billion, with the round nearing its close. The move directly resets its listing timetable: under Hong Kong Exchange rules, a company that has filed its application and then completes a new financing must, absent a specific waiver, wait 120 days after closing before listing. If the round closes at the end of October 2026, the earliest possible listing is late February 2027, ending hopes of a 2026 debut. In July 2026 the company had announced preparations for a Hong Kong IPO with a cornerstone target range of HKD 40–50 billion, a window now closed. Operationally, revenue exceeded RMB 1 billion in 2025, humanoid shipments rank among the world's highest, and products are deployed in hospital guidance and factory assembly, with the company also supplying core components and an embodied-AI software framework. Founded in Shanghai in February 2023, it is chaired by Deng Taihua, with Peng Zhihui as co-founder and CTO.
Capital milestoneZhangtongshe / Ruicaijing · 16 Oct
On 16 October, FAIRINO, based in Suzhou, announced back-to-back Series C+ and D rounds totalling nearly RMB 1 billion. The two rounds were led respectively by Meituan Longzhu and Meituan, with Yongxin Capital, Shishan II and Su Venture Capital participating, while existing shareholders Guolian Minsheng Capital, Su Chantou and Qingliu Capital followed on. Proceeds will mainly fund fixed-capital investment in core component capacity and the expansion of industrial embodied-AI production and R&D, covering the ART industrial humanoid force-controlled arm and the AIRLab tooling platform. Founded in Suzhou in April 2019, the company stands out for making its own controllers, encoders, servo drives, harmonic reducers, servo motors and brakes. For 2026 it expects 18,000 orders globally, up 68%, with overseas volumes up 87%; Europe accounts for 46% of overseas sales, Japan 16%, North America 13% and Southeast Asia 11%. The ART arm launched in August offers seven degrees of freedom and ±0.03 mm repeatability.
Capital milestoneChuanguan News / Cover News · 16 Oct
On 16 October, EIR disclosed a pre-Series A of more than RMB 100 million and said a further round had already begun. Investors include the Chengdu Future Industry Angel Fund, Deyang Holding Group, the Kaizhou New City industrial fund and the Zhongjiang High-Tech Zone fund, alongside support from the Sichuan Pilot-Scale R&D Company, forming a provincial-municipal-district coalition. Proceeds will fund iteration of the AIQ emotional-interaction humanoid, a 10,000-unit production base in Deyang and a national sales system. Founded in September 2024 in Chengdu and incubated by the University of Electronic Science and Technology robotics centre, the company closed a seed round in the tens of millions of renminbi in 2025. Its "Emotion Native" thesis rests on the EIR Harness architecture orchestrating perception, memory, interaction, navigation and motion; the face uses a 3D ultra-short-throw projection, a first in China. AIQ starts at RMB 98,000 in interaction and motion-control versions, with about 40% of structural parts in magnesium alloy.
Main funding events of the period (15–19 October)
Funding tableCompilation of public reports · 19 Oct
| Company | Round | Amount | Lead / main investors | Date | Segment |
| AGIBOT | New round | RMB 1–1.5 bn (approx. RMB 50 bn post-money) | Undisclosed | 17 Oct | Humanoids |
| FAIRINO | C+ / D | Nearly RMB 1 bn combined | Meituan Longzhu, Meituan | 16 Oct | Cobots / force-controlled arms |
| EIR | Pre-Series A | Over RMB 100 m | Chengdu Future Industry Angel Fund, Deyang Holding, Kaizhou New City, Zhongjiang | 16 Oct | Interactive humanoids |
| Gewu Tongzhi | Angel | Nearly RMB 100 m | Hongfu Asset Management, Peak Investment, Hangzhou Capital, Hangzhou Talent Group | 18 Oct | Manipulation AI |
| TigerX | Angel | Tens of millions of RMB | Puhua Capital, Hengze Capital | 16 Oct | Multisensory world models |
| Minerva Humanoids | Pre-seed | Approx. USD 10 m | General Catalyst | 16 Oct | Humanoids for hazardous work |
| Origin | Cumulative | USD 20 m | Accel, Zacua Ventures | 16 Oct | Interior-fit-out robots |
The amount structure is clearly bipolar:
two deals above RMB 100 million (AGIBOT, FAIRINO) account for roughly RMB 2 billion, the bulk of disclosed capital in the window; at the other end, two angel rounds of nearly RMB 100 million and tens of millions (Gewu Tongzhi, TigerX) target the model and perception layers rather than hardware. The two overseas deals focus on hazardous-work substitution and interior fit-out, at the tens-of-millions-of-dollars scale.
Market ceilingThe Paper / TMTPost · 16 Oct
Using the USD 1 billion threshold, China now has 28 embodied-AI unicorns worth a combined RMB 350 billion-plus, spanning robot bodies, embodied brains (foundation models) and dexterous hands. Integrated body-and-brain players are the most numerous (10 companies, about RMB 135.7 billion), with Galaxea, Galbot and AI² Robotics at RMB 20 billion each and AGIBOT around RMB 18 billion. The brain segment has only six unicorns but RMB 80.37 billion, an average of about RMB 13.4 billion, above the body-makers' average; dexterous hands count just four companies at roughly RMB 50 billion, of which Lingxin Qiaoshou alone is RMB 21 billion. Listing progress diverges: Mech-Mind has passed the Hong Kong Exchange hearing, Leju Robotics and DEEP Robotics are in the exchange's questioning phase, X Square Robot has filed confidentially, and Galaxea and EngineAI have completed incorporation as joint-stock companies. Valuation ramps are accelerating: Sudo Robotics went from first funding to a RMB 14 billion valuation in about six months, while Kunlunhang, Zhijian Dynamics and AGILINK crossed the unicorn line in under six months, versus eight to ten years for Unitree, Pudu, Fourier and BrainCo. Wu Zuxuan of Fudan's embodied-AI institute sums up the field as "bodies slightly ahead, brains broadly in step".
Early stageTechNode / IT Home · 18 Oct
Hangzhou-based robot-AI startup Gewu Tongzhi has closed an angel round of nearly RMB 100 million from Hongfu Asset Management, Peak Investment, Hangzhou Capital and Hangzhou Talent Group. Founded in 2026, it develops AI for dexterous manipulation, combining world models with learning from physical interaction. Founder and CEO Cai Jia previously worked in Huawei's research labs and automotive business, contributing to its ADS driving systems. Proceeds will fund model training, hardware development, hiring and real-world data collection. Commercially, the company is running proofs of concept with clients in textiles, optical modules and automotive components, with first delivery expected by the end of 2026. Its positioning is notable: rather than building a body, it packages "dexterous hand + manipulation model + field data" into a deliverable capability.
Technology routeCNR / Jilin University / Economic Observer · 16 Oct
The biomimetic hand Y-Hand M2, developed by Professor Ren Lei's team at Jilin University, has fully disclosed its internal biomimetic architecture for the first time. With 38 degrees of freedom, it is the highest-DoF biomimetic dexterous hand among publicly documented designs, completing all 33 categories of human-like manipulation with a 96% success rate, close to human level. Its foundation is the "biomimetic tension-compression robot" theory Ren built after more than 20 years of research on the neuromuscular-skeletal system: soft tensile tissue and hard compressive tissue form a spatial topology that breaks with rigid hinge kinematics and reproduces the six natural degrees of freedom of a biological joint. By subsystem: the mechanical structure uses biomimetic bone and cartilage, 3D joint surfaces and metamaterial-woven ligaments; actuation uses proprietary YQ Motors and YQ Muscles artificial-muscle modules; transmission relies on a tendon-sheath network enabling cross-joint power transfer; sensing draws on multi-source encoders and force sensors; control uses a spiking neural network with visuo-tactile fusion. Industrialisation sits with Yuequan, which already produces and ships at scale. The approach is a clear departure from mainstream rigid hands built on "motor + reducer + linkage or tendon", at the cost of harder engineering consistency and cost control.
Product offerEconomic Daily / Securities Times / OFweek · 18 Oct
Leju Robotics has introduced an industrial solution built around a "universal humanoid body + universal dexterous hand", whose SG100 variable-configuration hand is the key variable. Specifications: four fingers, 11 active degrees of freedom, a pinky rotating 180°, 7+N grasp modes (two-finger pinch, internal spread, enveloping grasp), at least 24 kg hook payload per hand and 0.03 mm repeatability — enough for both picking screws and carrying totes without changing end effectors. The body is the KUAVO 5W, developed from more than 100 industrial scenarios, with nearly 1,000 hours of failure-free continuous operation and EMC and 3B certification. On the software side, the Taskor platform wraps perception, mobility, grasping and manipulation into drag-and-drop modules, which the company says cuts some industrial deployments from months to hours. In the field, Leju and Zhaofeng Electromechanical deployed 15 humanoids in continuous operation on an automotive-parts line: more than ten action sequences, reliable grasping of parts with 80–120 mm inner diameters, a loading cycle under 40 seconds, 15 workstations and more than ten part types covered, 98% success, over two months without failure and about 1,000 cumulative hours producing more than 60,000 parts; the resulting real-world dataset was donated to the OpenLET community of the Open Atom Foundation.
Commercial watchChina Industry News / Sina Finance / NBD · 15 Oct
Several dexterous-hand makers are moving to "volume on the line" without reaching a profitability inflection. Ni Hualiang, founder of OYMotion, says several thousand-unit orders are in delivery this year, validating the production system; Wang Peng, founder of CasiaHand Robotics, reports a bookings value above RMB 100 million in the first half, while noting that stability and reliability at scale remain the biggest obstacle. Inspire Robots is blunter: its miniature servo-cylinder business is profitable, while the dexterous-hand business remains in an investment phase; most players still grow on funding. Wang Peng of the Beijing Academy of Social Sciences proposes a three-tier framework: production capability (whether the transmission route is engineered, core components are in-house, and assembly and testing are automated and standardised), profit outlook (whether the customer mix shifts from research to industry, and the marginal cost curve), and durable moats (proprietary tactile datasets and operational models). Divergence also shows in architectures: tendon hands suit generic grasping, linkage hands suit logistics sorting and machine tending, and direct-drive hands suit research, education and special applications; with no convergence yet, customisation remains unavoidable near term.
Cost structureNanfang Daily / Guangdong Provincial Government · 16 Oct
In the service-robot zone of the 140th Canton Fair, OYMotion showed its ROHand dexterous hand, the OpenArm open-source dual-arm robot and an industrial three-finger gripper. In live demos the hand detects forces as low as 0.1 N; the more consequential disclosure is on cost: the company's in-house motor has fallen from RMB 4,800 per unit to the low hundreds, bringing the motor cost of a complete hand below RMB 1,000. On capacity, it claims annual output at the 10,000-unit scale, with several thousand-unit orders booked this year and overseas demand doubling. Its market read is worth recording: European and North American customers have solid automation foundations but early-stage embodied AI, so they prefer proofs of concept, trials and small batches, with demand concentrated in research and education; Chinese manufacturers push dexterous hands into industrial settings more aggressively; Japan and Korea are growing fast. At the same show, Dobot presented "Lumeng", billed as the world's first all-terrain embodied humanoid, able to cross sand and pebbles with multimodal emotion perception and 3D spatial understanding.
LaunchGuandian / Leaderobot · 16 Oct
On 16 October Beijing time, Faraday Future launched the FF Nova humanoid at USD 2,490, alongside FF Nova Mini for physical-AI education. Sales roll out in three stages: a limited 100 units for developers and educational institutions, with 20 delivered immediately; full release to all developers and schools; then a consumer version. The launch also introduced a Physical AI Spark course, a NOVA reinforcement-learning developer programme and community recruitment. The price pushes the bipedal humanoid anchor lower still: for comparison, Songyan Dynamics' Xiaobumi is RMB 9,998, Unitree's R1 moved from RMB 39,900 to RMB 29,900, and Astribot's T1 starts at RMB 89,900. One distinction matters: the USD 2,490 tag applies to a developer and education edition whose functional scope is not comparable with an industrial humanoid; the commercial significance lies in lowering the entry point for a programmable humanoid platform to consumer-electronics pricing.
LaunchThe Terminal / The Robot Report · 15 Oct
On 15 October, Agility Robotics said Nyobolt, a Cambridge-based UK battery company, supplies the fast-charging system for its Digit 5, which the company says can deliver more than 20 productive hours in a 24-hour day. The arithmetic rests on a 10:1 run-to-charge ratio — 90 minutes working for nine minutes charging: a 99-minute cycle fits 14.5 times into a day, yielding about 21.8 hours of runtime and 2.2 hours of charging, a theoretical ceiling that assumes no waiting for a dock, no faults, no handovers and no maintenance. By comparison, the Digit 4 ran at 2:1, or about 16 hours a day. The commercial implication is more direct: based on a disclosed RaaS fee of about USD 8,500 a month plus a USD 25,000 deployment fee, 20 hours a day works out to about USD 14.17 per hour, versus USD 17.71 at 16 hours — at a fixed monthly fee, each additional productive hour cuts the hourly cost by roughly 20%. Direct purchase is about USD 200,000, plus USD 20,000 deployment and about USD 36,000 a year for software and maintenance. Battery capacity in kWh is undisclosed, so charging power cannot be derived; early-access units are expected in the first half of 2027.
Orders and deliveriesShenzhen TV / Economic Daily · 18 Oct
Under Frost & Sullivan's 2026 global humanoid market report, UBTECH ranks first worldwide with RMB 620 million in humanoid revenue in the first half and a 16.5% share, on 16,123 units sold. Total first-half revenue reached RMB 1.269 billion, up 104.2%; orders focus on aerospace manufacturing, automotive and intelligent manufacturing, with more than RMB 1 billion in new orders, including three worth over RMB 100 million each; Airbus ordered an initial 100 Walker S2 units, the first humanoid order in global aerospace manufacturing. On capacity, the Liuzhou super-factory, which began production on 12 September, targets more than 10,000 units a year, rolling one unit off the line every ten minutes. On ecosystem, a strategic agreement with FAW-Volkswagen on logistics was signed on 8 October, followed on 9 October by one with Xiamen C&D to form a joint venture serving as its Fujian regional headquarters, targeting cumulative sales of at least 10,000 humanoids by 2030. Its proprietary dataset holds about 11 million entries, more than 80% from industrial scenarios.
Exports and channels21st Century Business Herald / Tencent News · 17 Oct
The service-robot zone at the 140th Canton Fair has become the busiest part of the show, with humanoids, quadruped robots and dexterous hands taking turns, and some exhibits winning orders on the spot. Buying behaviour is changing structurally: at earlier editions visitors treated Chinese robots with curiosity and asked functional questions, whereas today more arrive with specific scenario requirements, and demand is segmenting by market. According to Huang Jun, who leads embodied-AI solutions at OYMotion: European and North American customers have strong automation foundations but face an early-stage embodied-AI market, so they prefer proofs of concept, trials and small batches, with demand concentrated in research and education; Chinese manufacturers push dexterous hands into factories more aggressively; Japan and Korea are growing quickly. On supply chains, Shenzhen's Huaqiangbei clusters more than 115,000 businesses spanning chips, sensors and full-system integration. The reporters' conclusion: what gets exported is not just a machine but a full delivery capability, plus the production capacity that travels with the orders; the power to decide product iteration is shifting from R&D to the shop floor, and the question is no longer "who are we" but "how fast and in what volume".
Sector dataCS.com.cn / China Securities Journal · 19 Oct
Real-robot data, body-free collection data and simulated data form the three main "nutrients" of embodied-AI training, each with its own production mode and limits; the industry is scaling volumes fast, but data quality and evaluation standards are seen as the decisive bottleneck. On infrastructure, more than 70 embodied-AI training grounds are operational in China, with more than 40 under construction or planned. On the gap, a CAICT report finds that a "ChatGPT moment" for embodied AI requires at least tens of millions of hours of real-world data, while the world's available high-quality real data amounts to only hundreds of thousands to a million hours — roughly an order of magnitude short. Capacity figures are emerging: the Beijing Humanoid Robot Innovation Center reaches 180,000 hours a year through matrix-style collection and has already delivered nearly 30,000 hours of high-quality data; Noitom Robotics supplies three categories, with up to 300,000 hours a year of lightweight ITW data, about 100,000 hours of MOV large-space interaction data and about 50,000 hours of OM multimodal fine-manipulation data. On demand, data centres bought 3,181 humanoids in 2025 out of 18,300 units shipped, or 17.4%, and global data-centre demand is expected to grow 140% in 2026 to 7,635 units.
Training groundsChina Fortune Net / China.com · 19 Oct
On 15 October, Hong Kong, China opened its first "International Embodied-AI Application Training Ground", where companies sent their robots as "star pupils" to perform specific tasks at different posts. On display: AI² Robotics' AlphaBot operating a drinks machine, Galbot's robot picking and selling goods from a shelf, and Robot Era's M7 identifying a parcel label and flipping it face-up with its dexterous hand. The mechanism: training, validation and optimisation in real scenarios generate data that feeds model training. Such sites arise from cooperation between local governments and industry to let robots "learn" in industrial or consumer settings; Beijing, Shanghai and Shenzhen have already built collection centres recruiting robot "pupils" guided by human operators. For observers, the relevant metric is not the number of sites but the variety of replicable tasks per scenario and the efficiency of data return — task homogeneity across sites remains a marked problem.
Cost structureFourth Wave / CAICT training-ground report · 19 Oct
According to CAICT's 2026 report on training grounds and several estimates, training a general embodied model requires at least tens of millions of hours of real interaction data, while the world's high-quality real-robot corpus is only in the millions of hours; expert teleoperation typically yields only tens of hours a day. The industry has therefore settled on a three-tier data pyramid: pre-training on human video and internet data for general representations, mid-training in simulation for the long tail, and fine-tuning on real data for deployment accuracy. The cost gradient is sharp: real-robot teleoperation at RMB 500–1,000 per hour (highest precision, perfect alignment with the target body, lowest sim-to-real risk, but slow to scale); body-free collection at RMB 100–600 (no force information, requires retargeting); simulation at RMB 20–50 (very broad coverage but with transfer bias); human-video distillation from very low cost up to RMB 100 (visual information only, used for cold start). On policy, in June 2026 the Ministry of Industry and Information Technology and SASAC launched a dedicated real-world training programme anchored in actual production and service posts; Mifei's lab in Zhangjiang is one such site, where an operator must repeat gestures as simple as picking up a bottle hundreds of times.
Capacity and ordersXinhua Finance · investor-relations record · 15 Oct
Leisai's 15 October investor-relations record details its robotics business. The portfolio spans frameless torque motors, planetary and harmonic joint modules, integrated wheel modules, coreless motors and dexterous hands at several DoF levels, all already in volume delivery. The commercial model has three tracks: supplying components and solutions, co-developing joint modules and hands with major accounts, and contract assembly. Nearly 300 customers at home and abroad, with the company claiming a place inside more than 80% of humanoid OEMs' supply systems. Order and capacity figures are specific: over one million frameless torque motors ordered for the year, 800,000–900,000 deliveries expected, current monthly capacity of 150,000 units, with automated lines being expanded toward three million a year; more than 60,000 joint modules on order, with deliveries accelerating. Financially, robotics should account for about a tenth of revenue this year, with RMB 200–300 million in related orders. On dexterous hands, it is selling linkage-based low-DoF units for cost-sensitive volume, developing a direct-drive high-DoF hand due for launch shortly, and staying cautious on tendon drives because of creep and durability issues.
Supply rhythmEastmoney / Xinjian inquiry-letter reply · 17 Oct
A clear gap has opened between supplier shipments and final assembly. Supply-chain reports indicate that weekly deliveries from core suppliers to Tesla reached roughly 1,200–1,500 robot-equivalents in October, while Tesla's own factory output has only just climbed to 400–500 units a week, up from 100–200 in early September. Suppliers are therefore running one to two months ahead of line output: the chain is stocking, the line is ramping, and the volume inflection lies ahead. On technology, Xinjian Transmission argues in its reply that heavy-load humanoid joints must use planetary roller screws, the optimum reached by leading OEMs through simulation and engineering validation, with no substitute in the near term — so the benefit concentrates in linear actuators and these screws. The domestic chain's order of certainty runs: actuator assemblies, planetary roller screws, reducer gears, dexterous-hand parts, then sensors — the first two leading, sensors trailing. On equipment, Pacific Precision Forging confirms it has developed and delivered samples of cycloidal and planetary reducers and planetary roller screws, including to overseas customers, with volume lines being commissioned. These weekly figures come from supply-chain surveys and media synthesis, not from Tesla disclosures.
Foreign entrantsQichacha / World Robot Conference · 18 Oct
Schaeffler established a dedicated embodied-robotics company in Taicang, Jiangsu, on 27 February 2026, focused on humanoid core components and key subsystems: rolling bearings, roller screws, precision reducers, motors and sensors, plus integrated solutions for rotary drive, linear motion, thermal management and electric drive. Exhibiting at the World Robot Conference for the first time, it showed these components and subsystems, and has secured a nomination from a local humanoid customer, with production to follow. On ecosystem, it signed a strategic agreement with ROKAE on joint modules to close a "R&D–supply–deployment" loop, and another with Leju Robotics, naming it its China embodied-AI partner. The sector-wide implication is notable: in 2026 Bosch, Schaeffler, Valeo, Aptiv and ZF are all moving aggressively into embodied AI through dedicated companies, technology transfer and cross-sector alliances — bringing not only capacity but automotive-grade quality systems and global customer networks.
Import substitutionSector synthesis (public sources and broker notes) · 19 Oct
In a humanoid's bill of materials, reducers, motors and screws together account for about 59%, against 8–12% for the compute unit, battery and power electronics — value sits in mechanical transmission, not the "brain". Reducers are the most thoroughly localised link: Leaderdrive holds over 60% of the Chinese market and ranks second globally, with 292,000 units shipped in 2025, capacity of 70,000 units a month as of July 2026, a year-end target of 100,000–120,000 and orders booked into 2027; Shuanghuan Driveline has lifted its RV share from 9% in 2020–2021 to 15%, targeting over RMB 1.5 billion in robotics orders in 2026 and about RMB 2.8 billion in humanoid-related orders; Zhongda Leader is the only maker producing harmonic, RV and planetary families in volume, supplying about 63% of Unitree's planetary reducers. Substitution has crushed prices: a same-spec RV reducer fell from RMB 1,280 in 2023 to RMB 410 in 2026, down 67.9%. Screws remain the weak link: domestic content in planetary roller screws is only about 29%, German and Japanese firms hold over 80% of the high end, yields run around 60%, and each robot needs about 14 units. On structures, Everwin Precision delivered more than 1.1 million precision humanoid components in January–August 2026, against about 690,000 in all of 2025, with RMB 243 million of first-half revenue in embodied AI and emerging hardware, up 537.64%.
Policy and complianceMOFCOM Announcement No. 48 / Xinhua · 16 Oct
On 16 October, China's Ministry of Commerce and Ministry of Science and Technology issued Announcement No. 48 of 2026, opening a public consultation on revising the catalogue of technologies prohibited or restricted from export, with interim pre-export registration for the technologies concerned. The revision adds 40 critical technologies across artificial intelligence, biomedicine and photovoltaics — 14 to be prohibited and 26 restricted; the AI section covers 23 technologies related to large models, AI agents, compute clusters and information processing. Three facilitations are spelled out: technologies already in the public domain through open source or published papers are excluded; restricted technologies intended for open source qualify for automatic licensing; and self-built overseas compute centres, overseas chip tape-out, joint R&D using overseas resources and participation in international standards qualify for automatic licensing, effectively "file and go". The boundary matters: controls target underlying technology, not products or services — absent an actual transfer of model weights, core training data, source code, key algorithm documentation, compute-system design or sensitive agent implementations, companies may still compete overseas through APIs, SaaS, model applications, industry solutions and hardware. The announcement adds that most listed technologies carry explicit technical thresholds and that products manufactured using them are outside scope.
Market structureFrost & Sullivan 2026 report · 18 Oct
Frost & Sullivan's report finds that 36,000 humanoids were sold worldwide in the first half of 2026, exceeding the 35,000 sold in all of 2025, generating USD 540 million in revenue, also above 2025's USD 500 million — six months matching a full year. The striking finding is structural: by revenue, the global top five are all Chinese and hold 57.4% combined. The ranking: UBTECH RMB 620 million (16.5%, 16,123 units); AGIBOT RMB 610 million (16.1%, 7,850); Unitree RMB 580 million (15.3%, 4,900); Galbot RMB 210 million (5.5%, 308); Leju RMB 150 million (4.0%, 3,540). Revenue and unit rankings diverge: UBTECH sells the most, Unitree commands the highest unit price, Galbot pursues a high-technology, low-volume route. The report judges that Tesla Optimus, Figure AI and Agility have accumulated hardware and base-algorithm expertise, but their products remain largely in prototype iteration and small lab pilots, with deployment mostly internal, keeping them out of the top five. Chinese advantages are attributed to a complete manufacturing system, a dense supply chain and abundant downstream scenarios — in practice, delivery speed. The report adds a caution: applications split into entertainment, research and education, data collection and production work, and the share actually working on production lines remains limited.
Humanoids abroadHumanoids Daily · 18 Oct
On 18 October, Elon Musk said on X, replying to a discussion of the Roadster's commercial importance, that "Optimus and Cybercab will be the vast majority of Tesla's value", linking the judgment to the potential scale of AI, robotics, solar power and space transport; he offered no valuation split, Optimus sales forecast or new delivery commitment. On capacity, an independent observation emerged: Joe Tegtmeyer, who tracks the Texas build by drone, said on 17 October that Tesla appeared to be completing the Optimus factory building, roof included, before installing major equipment, unlike Giga Texas where lines went in before the building was finished; he added that major equipment deliveries had not yet occurred. This is an outside observer's account, not confirmation of procurement or a production timetable, and the site is separate from existing Optimus production in Fremont. Official commentary remains that Optimus has roughly 10,000 unique parts, that "the initial portion of the S-curve will be quite flat and long", and that 2026 output goes first to Tesla's own factories for training and data collection. Sector consensus holds that fingertip tactile sensors and the assembly of micro screws and actuators remain the biggest bottleneck, with many steps still manual.
Humanoids abroadHuanqiu / Engadget / Gulf Times · 18 Oct
OpenAI CEO Sam Altman confirmed on a podcast that the company will build its own humanoid robot. His argument is about form: "the world is designed to a large extent for humans", with doors, keyboards and tools built around a particular body plan, so a robot moving through the same spaces should share it. Asked whether OpenAI would really build one, he replied: "We will." He stressed, however, that aesthetic details "matter far less than actually making the robot's brain work" — a robot that cannot use its limbs gains nothing from having them. On timing, Altman said "I think we'll have a great demo in 2027", while placing real public use years away, with no disclosure of the demo's content, the robot's appearance or any public access date. Job listings show teams forming across motion control, state estimation, actuator control, whole-body behaviour and physical-AI training infrastructure, along with in-house mechatronic actuators integrating motor, transmission, sensing and thermal management, plus roles in manufacturing and supply chain — hardware is not being marginalised. Its earlier Dactyl project trained a hand in simulation to solve a Rubik's Cube, already exposing the sim-to-real gap.