I. Computing Power Infrastructure: The "New Infrastructure" Boom Amid Shortages

If artificial intelligence is the engine of industrial transformation, then computing power is its fuel. And right now, that fuel has never been more expensive or more scarce.

1. GPUs: In Desperate Shortage, Leasing Models Explode

At the exhibition area, booths of computing power service providers such as Bit Computing, Jie Computing, Tenyun Computing, and Dingsuan Intelligence were crowded with visitors. Their display boards prominently featured sought-after models like "B200/B300/H100/H200/L40s" and "4090/5090/A100," while slogans such as "No Refurbished/No Gray Market," "Authenticity Traceable," and "Ready for Immediate Shipment" hinted at the market’s chaos and anxiety.

"Lead times for high-end GPUs have now stretched to over six months, and prices change almost daily," one service provider told us. This supply-demand imbalance has directly fueled the explosive growth of the GPU leasing market.

Tenyun Computing showcased its "HAT Intelligent Cloud" platform, which offers "elastic computing power available on demand." Supporting bare metal, container clouds, and various cloud computing models, the platform helps AI projects go live quickly and handle fluctuating computational needs with agility. Its proprietary "IRIS inference acceleration" service suite promises to significantly boost throughput and computing efficiency without sacrificing accuracy.

Jie Computing promoted itself with slogans like "Lowest Prices Nationwide" and offered on-demand leasing with elastic scaling. It also provides GPU repair services, even posting a detailed repair timeline—from five business days for ECC errors to fifteen for motherboard issues—highlighting the growing aftermarket service ecosystem surrounding computing hardware.

2. AI Servers: Performance Race Escalates from General to Specialized

The GPU shortage hasn’t only driven up prices; it has also triggered a comprehensive upgrade in the AI server market.

Roycom unveiled its RW7260-165-SP high-performance AI computing server. Built on Intel’s 6th-gen Xeon 6700/6500 series processors, it supports up to eight full-height, full-length, 3.5-slot-wide GPU cards and can expand memory to 8TB. Its 7U rack design and independent CPU/GPU airflow architecture provide a highly reliable platform for compute-intensive tasks like AI training and inference.

Omnisk brought the NVIDIA DGX Spark, touted as "a Grace Blackwell AI computer for your desk." With 128GB of unified system memory, 1,000 AI TOPS of performance, and the ability to link two units to support models with up to 405 billion parameters, this desktop-class device is positioned as an ideal solution for smaller R&D teams.

TaoCloud focused on the AI cloud infrastructure layer, presenting its all-flash SD52.0 storage system, including FASS distributed redundant storage and FOSS distributed redundant object storage. Its messaging—"giving wings to computing power with storage"—underscores the critical role of high-speed storage in large-scale AI model training.

3. Storage and Memory: Definite Winners in a Rising Price Cycle

As large language models cross the trillion-parameter threshold, the importance of storage and memory has reached unprecedented heights. At the expo, Seagate showcased its Exos X enterprise hard drives, with 32TB capacity as a highlight; UGREEN promoted its DX series NAS devices, emphasizing "smart, secure, and reliable" operations.

Of particular note, Helison (Beijing) introduced its next-generation AI storage server, featuring fully on-premises deployment, innovative AI-powered storage, and a self-developed AI document management system. Its products support seamless integration with external NAS devices, enabling stored documents to be incorporated into an AI knowledge base—an attractive proposition for government and enterprise clients seeking localized, secure solutions.

On the services and recycling front, Dingsuan Intelligence operates across the full lifecycle of computing hardware—from selling servers and GPUs to providing chip-level repair services. This "hardware delivery + chip-level repair" integrated model reflects a pragmatic reality: in today’s market, maximizing the utility of every available chip is essential.

II. Robot Components: Domestic Substitution Accelerates with Key Technological Breakthroughs

If computing power is the AI "brain," then robot components are its "muscles" and "skeleton." At this conference, China’s robotics components industry demonstrated a landscape of rapid advancement, spanning actuators, bearings, bionic materials, and controllers.

1. Actuators and Drives: Rapid Rise in Domestic Manufacturing

Derby New Power exhibited two core products: the DB-T-4801-700 linear shifter (rated torque of 700 N·m) and the DB-R-6001-30 planetary joint module (peak torque of 500 N·m). Both highlight high rigidity, strong overload capacity, and compact structure, making them ideal for large joints in humanoid robots, such as shoulders and knees.

Wanzhida Motor displayed both its "dexterous hand motor" (for small joints like fingers) and "joint torque motor" (for large joints), demonstrating a product portfolio that spans from precision micro-motors to high-torque actuators.

Luoyang Baist Precision Machinery, a high-tech enterprise from Luoyang, showcased its specialized bearings and RV reducers for humanoid robots. As one of the few domestic manufacturers capable of producing both CNC lathes and CNC grinders, the company already exports to markets including South Korea, India, and Japan.

2. Materials and Surface Treatments: Tackling Lightweighting and Wear Resistance

Xinghang Zhitai presented its "Titanium Alloy & Robot" theme, emphasizing the advantages of titanium alloy in robot components: lightweighting for enhanced agility, high strength for load-bearing and impact resistance, and excellent dimensional stability to maintain precision over extended use. Applications include underwater robots, deep-sea explorers, aerospace robotics, and explosion-proof equipment.

Guangdong Kinzan New Materials focused on coating technologies, offering PVD coating solutions for sectors ranging from construction and automotive to medical devices and consumer electronics. In robotics, its coating solutions for tools and components provide wear and corrosion resistance—critical for robot joints and transmission systems.

Shanghai Xingen Technology displayed its TPU films, targeting the "bionic skin/robot" market, offering new material choices for tactile sensing and aesthetic simulation in humanoid robots.

3. Sensors and Perception: Chinese Lidar Achieves Scale

Pacecat showcased its full-scenario lidar lineup, including the LDS-E210 series, touted as the smallest industrial 2D lidar; the LDS-S300 series, its best-selling navigation lidar; and the LDS-50C series for rotating robots, low-speed unmanned vehicles, and 3D measurement applications. Based in Zhejiang, the company has already supplied nearly a thousand clients globally and is expanding into 3D solid-state lidar technology.

Dongguan Wanxing Technology presented a range of smart knobs integrating TFT displays and UART communication. Targeting applications in medical devices, smart bathrooms, and smart washer-dryers, these components exemplify the trend toward miniaturization and integration in human-machine interfaces.

4. Controllers and the "Brain-Cerebellum" System: Domestic Solutions Achieve Mass Production

Suzhou Apache Robot Technology, a core supplier to Zhiyuan Robotics, presented its "brain and cerebellum" solution for embodied intelligence. With a decade of experience in industrial robot control, the company entered the embodied AI space in 2023 and has already shipped over 3,000 units of its KiWiBot series controllers to more than 50 industry clients.

Seavo launched its "brain and cerebellum fusion" embodied intelligence platform series, including HB02, HB03, HB05, and HB06 development platforms. Offering computing power ranging from 96 TOPS to 1,035 TOPS, and supporting Intel Core Ultra AI processors as well as Nvidia Jetson Thor, the platform provides a one-stop development solution for robotics companies.

Beijing Humanoid Robot Innovation Center, a joint venture established by four shareholders—Jingwei Electromechanical, UBTECH, Xiaomi, and E-Town Robot—positioned itself as China’s first full-stack embodied intelligence company. Its "Embodied Tiangong" general robot platform and "Huisi Kaiwu" general embodied intelligence platform aim to accelerate industry progress through open-source technology sharing.

III. Application Scenarios: From Mines to Homes, AI Accelerates Deployment

Enabled by advances in both computing power and components, AI and robotics are rapidly permeating diverse industries.

Xiaobu Ruying presented its smart mine inspection solution, combining AI, panoramic imaging, and 4D digital twin technology. The system uses two deployment modes—wearable smart helmets and rail-mounted robotic dogs—to achieve comprehensive, round-the-clock coverage. The core value lies in shifting equipment maintenance from "reactive" to "predictive," significantly reducing downtime and operational costs.

Andun Health, a pioneer in health risk prediction, unveiled its Andun Health Risk Alert Watch and AI Traditional Chinese Medicine Robot. Its health management system covers seven bodily systems and over 60 health indicators, leveraging proprietary "Tianhui AI Pulse Diagnosis Algorithm Model" and "Tianhui TCM Large Model" to monitor everything from heart rate and blood pressure to cancer risk and organ health. A prominent banner reading "40,000 Alerts in 10 Years" underscored AI’s real-world impact in chronic disease management and early risk detection.

Wyndme introduced its AI home companion robot, marketed as the "next-generation intelligent companion with an AI brain." Supporting voiceprint recognition, sound source localization, computer vision, long-term memory, and autonomous behavior, each robot develops its own "unique personality and temperament," moving AI beyond utility toward emotional companionship.

IV. Industry Outlook: At a Crossroads of Scarcity and Expansion

Surveying the entire conference, it’s clear that the AI and robotics industry is currently at a crossroads defined by both scarcity and growth:

On one hand, computing power shortages are reshaping the industry landscape. Core components—GPUs, HBM memory, high-capacity enterprise SSDs—are either in short supply or subject to constant price increases, with lead times continuously lengthening. This supply-demand imbalance has catalyzed the rise of the GPU leasing market and accelerated domestic substitution efforts. From AI servers and storage systems to liquid cooling and chip-level repairs, a vast "aftermarket" ecosystem centered on computing power is rapidly forming.

On the other hand, domestic manufacturing of robot core components is advancing at an accelerated pace. From actuators, reducers, and bearings to sensors and controllers, Chinese manufacturers have demonstrated full-chain capabilities from R&D to mass production. Key robotics firms like Zhiyuan Robotics are steadily increasing the share of domestic components in their supply chains, and controller manufacturers like Apache and Seavo have already achieved large-scale deliveries.

It’s also worth noting that capital is flowing rapidly into this space. KuiFei Space-Time Quantum Technology disclosed at the conference that it is seeking 1 billion yuan in Series A funding for its "humanoid physical therapy quantum robot" project, offering 5% equity—a post-money valuation of 20 billion yuan. This highlights the intense investor enthusiasm for the AI and robotics sector.

Conclusion

This March gathering in Beijing was both a showcase of technological achievement and a reflection of industry-wide anxieties. Computing power shortages are raising barriers to entry, but they are also spurring innovative service models. Domestic substitution in robotics components is accelerating, yet critical bottlenecks remain—particularly in high-end GPUs and advanced process chips.

Over the next three years, two strategic imperatives will define China’s AI industry: building out computing power infrastructure and deepening domestic capabilities in core robotics components. Companies that can successfully navigate this dual challenge—simultaneously addressing scarcity while scaling innovation—will be best positioned to lead in the next phase of global competition.

Intelligence leads the future, computing power reigns supreme, and the race to secure critical components is on. These were the clearest signals from this year’s conference—and the challenges that will shape the industry for years to come.