China Tech Today: Beijing Writes the Rulebook for Its Robot Race — November 12, 2026

Beijing has stopped cheering its robot industry from the sidelines and started writing its specification sheet. A new three-year plan sets hard numbers for how many robots the capital region should build, how long they must run, and how much weight they must carry — and on the same day, UBTECH landed its first 400-unit humanoid order and a MetaX-backed venture raised money to build the chips those robots will think with.

Beijing sets a 200,000-robot-a-year target — and grades robots like appliances

Beijing’s municipal government released its Three-Year Action Plan for the High-Quality Development of the Embodied AI Robotics Industry (2026–2028) on November 11, and it reads less like an industrial policy document and more like a product standard. By 2028 the city wants annual production capacity of 200,000 robots, at least five world-class humanoid manufacturers and ten domestically leading ones, with the Beijing-Tianjin-Hebei region supplying 25% of China’s embodied-robotics capacity and over 60% of key components sourced regionally.

The striking part is the performance targets: a complete robot must carry more than 40% of its own weight, run for more than four hours on a charge, go more than 2,000 hours between failures and last over five years. Component targets are just as specific: dexterous hands with a rated load life of one million cycles and a unit cost below 10,000 yuan, batteries at 350 watt-hours per kilogram with 1,500 charge cycles, and tactile sensors resolving force to 0.01 newtons. The plan also calls for more than one million hours of robot training data across at least 100 real-world scenarios, computing clusters above 50,000 petaflops, at least 50 demonstration scenarios, and a unique identification code for each robot for lifetime tracking and safety oversight.

Why it matters: this is the moment China’s humanoid race stops being judged on dance routines and starts being judged on durability. The plan also pushes vision-language-action models, world models and reinforcement learning — the software side — alongside the hardware. The State Council’s Development Research Center projects China’s embodied AI market will pass 400 billion yuan by 2030 and one trillion yuan by 2035. For Canada, where buyers are watching Chinese humanoids move from trade-show floors into warehouses and showrooms, Beijing just published the checklist those machines will be graded against. (The AI Insider; KIMKJ)

UBTECH lands its first 400-unit humanoid order — in car showrooms

UBTECH signed a strategic cooperation agreement on November 11 with Yongda Automobiles Services, a Hong Kong-listed luxury car dealer group: Yongda will buy 400 U1 Pro ultra-bionic humanoid robots, with delivery planned within this year. The robots will greet customers and promote products in Yongda’s showrooms; the pair also plan robot sales and leasing across automotive retail, plus a joint UBTECH experience centre in Shanghai.

Neither company disclosed a contract value, and no Hong Kong exchange filing had appeared as of November 11. At the U1 Pro’s list price of 169,800 yuan, 400 units would come to about 67.92 million yuan — a list-price estimate, not a disclosed deal value. UBTECH says this is the first large ultra-bionic humanoid order it has disclosed; its CFO has previously guided to 1,500–2,000 U1-series deliveries this year. The context matters: this follows China Telecom’s deal with UBTECH on November 9 targeting a million home robots in three years, and JD’s three-million-robot plan. The sales channel for humanoids is being built before the machines are proven in homes. (Servospace, citing The Paper and Sina Tech; RobotHOT)

A MetaX-backed venture raises $75 million to build robot brains that beat Nvidia’s Thor

Xixuan Chuangzhi, an embodied-AI edge-chip company jointly founded by UBTECH and GPU maker MetaX in May 2026, has completed a 500-million-yuan (about $75 million) angel round at a post-money valuation of 4 billion yuan. The round was co-led by GGV Capital and Matrix Partners China; UBTECH and MetaX each hold 35.01% of the venture.

The company’s first in-house chip is a robot-oriented “unified brain” SoC — a single chip combining the large-brain reasoning and small-brain control functions of a robot — with a stated research goal of comprehensively surpassing Nvidia’s Thor robotics chip. A validation platform for industry partners is planned for 2027, with the first SoC product and a full embodied-AI solution to follow in 2028. The honest caveat: a chip defined in 2026 and shipping in 2028 means no meaningful revenue for at least two years, and “surpass Thor” is an ambition, not a benchmark. But the structure is the story — a robot maker and a GPU maker now co-own the chip their industry depends on, rather than waiting for Nvidia or anyone else to sell it to them. (BigGo Finance; RobotHOT)

AgiBot open-sources LIFE, a framework for robots that learn from living

AgiBot has released LIFE — Lifelong Intelligence from Experience — an open-source framework that lets robots keep learning after deployment instead of freezing their skills at the factory. It has three parts: RoboBeats, a lightweight streaming model that watches a robot’s surroundings and decides when something is worth remembering; Plastic Memory, which stores raw footage and structured records so the robot can answer questions about what it has seen; and EDM (Explore–Distill–Mount), which turns solutions worked out on the job into validated, reusable tools and skills.

In AgiBot’s own tests, the framework scored 60.5% on SSS-Bench, a test of long-horizon robot memory, and a case study replaying 24 hours of video from a real robot showed it building usable memory from a day of experience. Those figures are company-reported and the case study is small, so treat them as a proof of concept rather than a proven capability. Still, the direction is important: the industry’s hardest problem is no longer making a robot do a task once, but making it remember how — and AgiBot just gave that layer away for free, a month after unveiling its GO-3 foundation model. (RobotToday; RobotHOT; arXiv paper)

Nexperia’s interim CEO says the split chipmaker could still reunite with China

Nexperia interim chief executive Stefan Tilger told Reuters on Tuesday that the Dutch chipmaker could still reunite with its Chinese operations after more than a year apart, and that taking an equity stake in the Chinese business was another possibility, provided European industrial interests were protected. If the split becomes permanent, he said, Chinese parent Wingtech Technology should be compensated.

Before the split, Nexperia made wafers in Europe and sent them to China for assembly, testing and packaging — chips that end up in cars and industrial equipment. Tilger said the Dutch business has since benefited from customers seeking a supply chain independent of China. The comment is a signpost, not a deal: no timetable, no terms, and a year of separate operations to unwind. But it is the first public signal from either side that the separation might not be final. (TechNode, citing Reuters)

UPDATE: Double 11 closes at RMB 1,776.4 billion as AI reshapes the shopping festival

The final Singles’ Day numbers put a scale on the AI-shopping story from yesterday’s brief: gross merchandise volume across online platforms during the 2026 Double 11 festival (October 7 to November 11) reached RMB 1,776.4 billion — about US$264.4 billion — according to third-party data monitor Syntun. E-commerce platforms accounted for RMB 1,679.0 billion, with Tmall first; instant retail added RMB 89.1 billion and community group buying RMB 8.3 billion.

The platforms competed on more than discounts: Taotian embedded AI across shopping and merchant operations, JD.com combined AI with offline stores and instant delivery, and Douyin pushed coupons through its Doubao AI app and the Hongguo short-drama app. Syntun’s own note: beyond GMV, whether products match demand, discounts are transparent and fulfilment is dependable “now matter just as much.” The festival’s eighteenth year suggests the metric that counts is shifting — from how much was sold to how much of it was sold by an AI. (TNGlobal / PR Newswire, citing Syntun)

The big picture

Today’s six stories are one story told six ways: China’s robot industry is moving from demonstration to specification. Beijing is writing the durability standards, UBTECH is signing real orders, a chip venture is being built to supply the brains, and AgiBot is open-sourcing the memory. The chip squeeze has not slowed any of this — it has pushed the whole supply chain inside China’s borders.

Watch next

Adaspace shares begin trading in Hong Kong on Friday, November 13 — the first market verdict on satellite AI computing. Alibaba’s Qwen4-Max preview ends at midnight on November 25: watch whether benchmarks or a general release appear before then.

Sources

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top