| A100 pcie40G - $ 200/card/month;BMS:8*Ascend 910B $ 3000/month 8*4090D - $ 1000/month 8*4090 - $ 1200/month 8*A100 pcie40G - $ 1800/month 8*A100 pcie80G - $ 4000/month 8*A100 nvlink80G - $ 5400/month 8*A800 nvlink80G - $ 5400/month 8*H20 - $ 6000/month 8*L20 - $ 2000/month 8*L40 - $ 2400/month 8*L40S - $ 2500/month 8*H100 - $ 13000/month 8*H200 - $ 20000/month 8*B200 - $ 30000/month 8*B300 - $ 48000/month |
From “Showmanship” to “Practicality”: A Panoramic View of China’s Explosive Growth in the AI Robotics IndustryIn 2026, China’s AI robotics industry is undergoing a profound transformation. The market is no longer a stage for “PPT robots” — mass production and commercial viability have become the hard metrics for survival. At the same time, the push for domestic substitution in computing power is accelerating, with homegrown GPU manufacturers challenging NVIDIA’s long‑standing dominance. These two major trends together form the core backdrop of China’s AI robotics industry today. I. Market Explosion: A Hundred‑Billion‑RMB赛道 Takes ShapeIn 2025, global humanoid robot shipments reached approximately 18,000 units, with sales of US$440 million, a year‑on‑year increase of over 508%. Chinese manufacturers performed particularly well — among the top nine global brands in sales volume, all were from China. According to the White Paper on the Development of the Humanoid Robot Industry (2025) released by the China Academy of Information and Communications Technology (CAICT), the global humanoid robot market reached RMB 17 billion in 2025, of which China accounted for RMB 8.5 billion, more than half. Entering 2026, growth has further accelerated. Estimates from Gaogong Robot, IDC, and other institutions suggest that domestic humanoid robot shipments for the full year could exceed 60,000 units, with the market size surpassing RMB 20 billion, and output is expected to grow by 94% year‑on‑year. IDC forecasts that China’s embodied AI robot market will exceed US$11 billion in 2026, and Chinese manufacturers will account for over 85% of global shipments of service and consumer robots. Data from the China Commercial Industry Research Institute also show that China’s embodied AI robot solutions market reached RMB 34.6 billion in 2025, is expected to reach RMB 44.9 billion in 2026, and will exceed RMB 81.5 billion by 2028. In terms of shipment structure, head concentration is very significant. According to Omdia, in 2025, Zhiyuan Robotics shipped more than 5,100 units, capturing 39% of the global market, followed by Unitree Robotics with 4,200 units; together they accounted for 71% of the global market. An IDC report similarly shows that Zhiyuan and Unitree each shipped about 5,000 units, placing them in the industry’s top tier with clear advantages in scaled delivery capabilities, product maturity, and application scenario coverage. The two companies are expected to capture nearly 80% of domestic humanoid robot shipments in 2026. The rapid growth of the market is driven by multiple factors. On the policy front, embodied AI has been written into the 15th Five‑Year Plan and was mentioned for the first time in the Government Work Report in 2025, marking its elevation to a national strategy. Leading industrial hubs such as Beijing, Shenzhen, and Shanghai responded quickly, forming a development pattern of “central planning, local implementation, and ecosystem collaboration.” On the demand side, an aging population and labor shortages are creating a rigid demand for substitution. The global humanoid robot market is expected to grow from RMB 20.6 billion in 2025 to RMB 1.68 trillion by 2035, with a compound annual growth rate of 49.2%. At the same time, China’s complete manufacturing system provides a unique “co‑located synergy” advantage for embodied AI. Core capabilities accumulated in the intelligent electric vehicle industry — perception systems, chips, motion control — are rapidly transferring to the robotics field. The Guangdong‑Hong Kong‑Macao Greater Bay Area has formed a “half‑hour supporting circle” from core components to complete machine manufacturing, significantly reducing product development iteration cycles and costs. II. Computing Power Foundation: A Collective Breakthrough by Domestic ChipsThe “intelligence” of AI robots relies heavily on underlying computing power, and GPUs are at the heart of that foundation. In 2025, the share of domestic chip makers in China’s AI accelerator server market climbed to about 41%, gradually narrowing NVIDIA’s lead. Total shipments of AI accelerator cards for the year were about 4 million units, of which NVIDIA shipped about 2.2 million (55% share), while domestic manufacturers shipped a combined 1.65 million (41% share). This change is a milestone — previously NVIDIA held a dominant 95% share in the Chinese market. Its lead is now shrinking significantly, driven by US export controls cutting off supply of high‑end AI chips and the increasing emphasis by the Chinese government and enterprises on self‑reliant supply chains. Among domestic players, Huawei Ascend leads by a large margin with about 812,000 shipments, accounting for nearly half of total domestic shipments. Alibaba Pingtouge (about 265,000 units), Baidu Kunlunxin, and Cambricon (each about 116,000 units) form the first domestic tier. Haiguang Information, Tianshu Zhixin, Muxi, and others occupy differentiated supplementary positions. In terms of product performance, domestic AI chips are rapidly approaching international advanced levels. Huawei Ascend 910B delivers 376 TFLOPS of FP16 compute power, while the Ascend 910C’s performance is comparable to NVIDIA’s H100; the Ascend 384 super node achieves 300 PFlops of cluster computing power. Cambricon’s Siyuan 590 delivers 345 TFLOPS of FP16, close to NVIDIA’s A100. DIGITIMES predicts that shipments of high‑end cloud AI accelerators from Chinese vendors will reach 2.123 million units in 2026, an annual increase of 136%, with Huawei’s market share exceeding 50% to remain the leader. Notably, the domestic large model DeepSeek has chosen Huawei’s Ascend 950PR as the computing base for its next‑generation flagship model, and the underlying code of its V4 model has been migrated from NVIDIA’s CUDA framework to Huawei’s proprietary CANN framework. This marks a shift for domestic computing power ecosystems from “alternative solutions” to “core choices.” Nevertheless, domestic chips still face three major challenges: dependence on advanced process nodes with room for improvement in yield, an unstable supply chain for HBM (high‑bandwidth memory), and the strong inertia of NVIDIA’s CUDA ecosystem, which constrains the developer ecosystem for domestic chips. III. Major Industry Players: Differentiated Competitive Paths Emerge1. Zhiyuan Robotics: Pioneer in Mass Production and Scenario DeploymentFounded in 2023, Zhiyuan Robotics has achieved a remarkable leap from startup to industry leader. With over 5,100 units shipped in 2025, it ranked first globally in shipment volume and exceeded RMB 1 billion in annual revenue, making it the fastest robotics company in China to reach that milestone. Zhiyuan’s core competitiveness lies in its “full‑chain self‑development closed loop” and “supply‑chain‑driven” model. Founder Deng Taihua has defined 2026 as the “first year of the deployment‑ready era” for embodied intelligence — the industry is no longer about who has a more impressive demo, but who can faster, more reliably, and more cheaply turn robots into repeatable, parallelizable, and scalable productivity units. In March 2026, Zhiyuan achieved the milestone of 10,000 robots produced. In April, it live‑streamed an eight‑hour real production line operation at the Longqi Technology Nanchang factory, where the Lingjing G2 worked as a “formal employee” on the line performing precision material handling, with an overall success rate exceeding 99.5%. Zhiyuan also released four robot platforms, six AI models, and seven productivity solutions, covering industrial manufacturing, commercial services, and special operations. 2. Unitree Robotics: Benchmark in Motion Control and Consumer MarketsFounded in 2016, Unitree Robotics is widely seen as the strongest contender for the “first humanoid robot stock” on the A‑share market. The company has completed 10 funding rounds, with a post‑C‑round valuation exceeding RMB 12 billion. It surpassed RMB 1 billion in revenue in 2025. In March 2026, the Shanghai Stock Exchange formally accepted its IPO application on the STAR Market, with plans to raise RMB 4.202 billion. Unitree’s core strengths are excellent motion performance and full self‑development of core components. Its robots are widely deployed in research, education, and consumer markets. Its approach emphasizes “showmanship” and individual skill demonstration, standing in contrast to Zhiyuan’s more “practical” style. At CES 2026 and MWC 2026, Unitree’s booth was set up as a “battle arena” for humanoid robots, fully demonstrating its accumulated expertise in motion control algorithms. According to IDC, Unitree’s shipments are about 5,000 units, placing it in the top tier of the industry. Its IPO proceeds will be mainly directed toward R&D of intelligent robot models (RMB 2.022 billion) and robot hardware (RMB 1.110 billion). After completion, it is expected to achieve an annual production capacity of 75,000 humanoid robots. 3. UBTECH Robotics: Full‑Scenario Player Listed in Hong KongUBTECH went public in Hong Kong at the end of 2023, with a market capitalization of nearly HK$50 billion. It has set a target of reaching an annual capacity of 10,000 industrial humanoid robots by 2026. The company possesses full‑stack self‑developed servo drives and the ROSA 2.0 robot operating system, with deployments across education, commercial, smart home, and other scenarios. Its Walker S2 industrial robot has already been deployed at Zeekr’s factory, increasing production efficiency by 30%. UBTECH’s differentiation lies in full‑scenario coverage and a strong patent portfolio, with a positioning偏向 commercial and industrial applications. The company recently acquired a 43% stake in Fenglong Corporation through a combination of block trade and tender offer, seen by the market as a key step toward a potential A‑share listing. 4. Galaxy General and Zhi Pingfang: Rising New ForcesGalaxy General focuses on unmanned retail scenarios. Its Galbot G1 has been operating 24/7 at several unmanned pharmacies in Beijing. The company has developed its proprietary GroceryVLA model, with synthetic simulation data accounting for over 99% of training data, making it a technology leader in simulation‑driven learning. In 2026, it completed a new funding round of RMB 2.5 billion, with investors including the National Artificial Intelligence Industry Fund and China Petrochemical Corporation (Sinopec). Zhi Pingfang focuses on a full‑body, full‑scene VLA (Vision‑Language‑Action) large model. It has originally developed the GOVLA (Global Omnibearing VLA) model, achieving for the first time unified output of whole‑body control and movement trajectory. It has also achieved impressive commercial results — a cooperation with HKC (Huike) to deploy over 1,000 robots within three years, with an order value of approximately RMB 500 million, setting a new industry record. The mean time between failures of its core components exceeds 50,000 hours, and it plans to reach an annual production capacity of 10,000 units in 2026. 5. Cross‑Industry Entrants: Automakers such as Chery Join the RaceAutomakers entering the humanoid robot field has become a new trend. Chery’s subsidiary Mojia Robot has begun direct‑to‑consumer sales of its full‑size humanoid robot “Moyin M1” at a price of RMB 285,800, making it the first humanoid robot in China developed by an automaker and sold directly to consumers. Mojia has completed the first batch of 220 humanoid robots for global delivery, with products deployed in over 100 real‑world application scenarios across more than 30 countries. On the overseas front, Tesla’s Optimus Gen 3 has entered real‑world factory deployment and is expected to begin mass production in 2026, further intensifying global competition. 6. Other Important PlayersLeju Robot, Jiasu Evolution, and Songyan Power have shipments of around 1,000 units each and are accelerating their expansion into key application scenarios. DeepRobotics (Yunshen Technology) has initiated an IPO辅导 process and launched the wheel‑legged quadruped robot “Shanmao M20” and the industry‑grade humanoid robot “DR02.” Topstar has been deeply involved in manufacturing for nearly 20 years, with full‑stack self‑development of core components and a “brain‑cerebellum‑body” three‑layer fusion architecture. Its X5 intelligent robot control platform is the first in China to support cloud‑edge‑device deployment. In addition, several companies such as Fourier Intelligence, Xinghai Tu, Paxini, and Zhongqing Robotics have valuations exceeding RMB 10 billion. Since the beginning of 2026, total investment and financing in China’s embodied AI field has exceeded RMB 20 billion. IV. Challenges and OutlookDespite the industry’s prosperity, challenges are equally evident. There are now more than 150 humanoid robot companies, more than half of which are startups or “cross‑industry” entrants. This raises concerns about overcrowding of similar products and a squeezing of R&D space. Core technology bottlenecks remain prominent: fine manipulation by dexterous hands, highly durable hardware, and the collection of AI training data are still industry‑wide challenges. The cost of key components remains high, with limited options, and the return on investment cycle is long. The localization rate of core components has increased to 75%, but full self‑reliance is still some distance away. Looking ahead to 2026–2028, the humanoid robot industry is at a critical validation window, moving from “prototype” to “mass production of tens of thousands of units.” 2026 has already been called the “first year of the deployment‑ready era,” with the industry theme shifting to “mass production and accelerated commercialization.” Whoever first achieves a closed loop for scaled deployment will take the commanding heights in this transformation. Declaration: This article is originally created by Shenzhen Cloud Engine - a cost-effective AI computing power service platform. For reprint, please indicate the source link:https://www.omniyq.com/en/sys-nd/471.html
|