The Zhejiang Humanoid Robotics Innovation Center, established in December 2023 in Ningbo with support from the municipal government and the team of Professor Shun Zhong from Zhejiang University, is at the forefront of commercial humanoid implementation. The NAVIAI line from this center, which utilizes 90% domestic components, is successfully operating in the facilities of SUPCON, GREE, JACK, and FOTILE.
In April 2026, a landmark contract was signed with Jack Technology to supply 2000 units of robots for clothing manufacturing, marking the first batch for global application in the apparel sector. These robots are capable of achieving 98% fabric separation, performing operations in 3 seconds, and ensuring alignment with an error margin of plus-minus 2 mm, approaching 80% of human efficiency.
The NAVIAI product line includes three types of form factors. Stationary robots are designed for clothing manufacturing workstations where movement is not required and feature a lower cost. Wheeled arm robots, such as NAVIAI-WA2, are equipped with an advanced bionic arm with 8 degrees of freedom, providing a re-positioning accuracy of plus-minus 0.02 mm, 6 hours of autonomy, and the ability to overcome obstacles up to 20 degrees. Bipedal robots are used for security patrolling, exhibition escort, and entertainment, as they can climb stairs and navigate obstacles.
The brain of these robots combines an autonomous cognitive system for planning and perception with a real-time control system analogous to the cerebellum. In May 2026, the RAM system was introduced—a three-dimensional spatial model developed jointly with CUHK and Zhejiang University—which supplements large models with accessible 3D knowledge to connect semantic reasoning and precise manipulation.
International expansion accelerated in 2026. The first major deployment occurred at the BEKO refrigerator factory in Turkey, where the European appliance manufacturer chose quality control as the initial scenario. The Zhejiang Center provided the hardware and software and participated in joint on-site engineering work. In March 2026, at the LogiMAT exhibition in Germany, the NAVIAI-WA2 operated continuously for three days without failure during a smart logistics demonstration involving KION Group, a giant in industrial vehicles and supply chain solutions. In July, the center participated in the IPLUSMOBOT event at AUTOMATE 2026, North America's largest robotics exhibition, where the full-sized bipedal NAVIAI-I2 was showcased. It performed precise sorting, autonomous walking, and voice interaction in fluent English with a response delay to audience questions of less than 3 seconds.
CTO Xu Xuecheng explained the company's strategy: first perfecting products within the internal operations of foreign groups, and then mass-producing them in Europe and North America. The company focuses on performing tedious and dangerous tasks rather than replacing jobs that employees fight for. NaviStore retail kiosks offer guide functions, cup dispensing, and ice cream preparation with a 3-second voice delay, 98% recognition, and 99.9% capture success. Xu predicts the emergence of home humanoids in about a decade. Chinese humanoid robotics is transforming from equipment demonstration to business model validation.
Moonshot AI's Kimi K3 model, the largest open-weight model with 2.8 trillion parameters, experienced service overload just two days after its release. This forced the company to suspend the acceptance of new consumer subscriptions. The number of neurons in this model is approximately three times greater than that of the previous generation, and the market demonstrated a supply-demand imbalance from the first hours of the launch.
Juan Zhengxin, head of corporate business for Kimi at Moonshot AI, described the reaction to the launch as unprecedented. The release of an open-weight model allowed any global small or medium-sized enterprise to use the capabilities of the advanced model for free. This move positions Kimi K3 as a direct challenge to DeepSeek's dominance in the open model ecosystem and demonstrates the ability of Chinese models to achieve global reach.
The impact on the foreign market was immediate and measurable: the release of Kimi K3 led to a fourfold increase in paying users abroad and a 400% growth in API revenue. Open-source models developed in China now account for 41% of global open model downloads, surpassing the United States in this metric. In global rankings of major large model usage, all six leading models belong to Chinese teams.
Chinese open-source models held the global open scale ceiling for 9 of the last 12 months, with the cumulative downloads of such models exceeding 10 billion. This data confirms that open source has become the most distinguishing feature of China's artificial intelligence industry, where accessibility and price-to-performance ratio are the main competitive advantages.
MiniMax also introduced a new model optimized for agent design scenarios, capable of performing complex tasks such as decomposition, code generation, and engineering collaboration. This model is designed for easy deployment on accessible computing infrastructure. Chen Tiantun, Director of International Business at MiniMax, noted that the company made specific algorithmic improvements for high-performance scenarios. As a result, MiniMax now serves over one million global corporate and developer clients, a fivefold increase in six months.
The joint releases by these companies illustrate a broader trend: Chinese AI companies are targeting global developer ecosystems through open-source distribution and affordable pricing. This contrasts with the subscription-based closed model strategy used by leading American AI companies.
Beyond the successes of individual companies, this phenomenon reflects structural changes in the Chinese AI industry. 248 AI chain companies published half-year reports by July 30th, with 159 of them reporting profit growth, accounting for over 60% of the total. The expansion of national intelligent computing infrastructure, localization of domestic software and hardware, and reduced computing costs have transformed AI from an expensive experimental technology into an accessible production tool. Key industries have achieved over 80% AI penetration, intelligent computing power has grown by 177% year-on-year, and the scale of the AI-related industry has exceeded one trillion yuan. The open-source strategy has transformed global competition in AI from a monopoly of a few giants into broader industrial competition, while AI sovereignty is becoming a policy priority in many countries, and open-source development is expanding the participants in model ecosystems across more languages, regions, and industries.
The global thermonuclear energy industry is attracting huge capital, reaching a cumulative funding of $14.24 billion, which is seven times higher than in 2021. Chinese scientist Wang Teng presented the progress of the EAST CRAFT and BEST projects, which bring China closer to launching the first thermonuclear energy power plant.
At the Conference on Future Trends on July 30, Wang Teng, a scientist from the Institute of Physical Sciences CAS in Hefei since the 1990s, stated that China's research in thermonuclear energy is world-leading, and the country is aiming to launch the first generation source based on nuclear fusion. Nuclear fusion requires plasma with a temperature of 100 to 300 million degrees and faces two main challenges: efficient ignition and long-term stable confinement.
Nuclear fusion is the ultimate goal of China's three-stage nuclear strategy, enshrined in the 15th Five-Year Plan, while the Atomic Energy Law will come into effect in January 2026. Global capital is actively flowing into this sector. According to the Thermonuclear Energy Industry Association's 2026 report, worldwide funding reached $14.242 billion, equivalent to approximately 96.4 billion yuan, which is seven times more than in 2021. In 2025, China invested about $4.7 billion, second only to the United States.
Since June 2026, eight Chinese companies involved in thermonuclear energy have completed funding rounds, including ENN Fusion, Zeta Fusion, Xinghe Fusion, Chaomag New Energy, and Beta Fusion. Xinghe Fusion announced a seed round of 830 million yuan in early July with participation from SAIC Capital, Shenzhen Capital Group, Hefei Innovation Investment, and Zijin Mining. Investor enthusiasm is explained by the fact that the power density of a thermonuclear reactor depends on the fourth power of the magnetic field intensity: doubling the magnetic field reduces the device volume by 30 times. Furthermore, the commercialization of nuclear fusion is accelerating due to the industrial production of high-temperature superconducting tapes and the application of artificial intelligence for predicting disruptions and optimizing configuration.
It is projected that investments in thermonuclear energy devices will amount to 20 billion yuan annually, providing a market potential of 50 billion yuan. Planned domestic thermonuclear energy projects exceed 150 billion yuan, and commercialization is expected around 2040. Hefei concentrates nearly 60 enterprises related to thermonuclear energy at all stages of the chain. China Fusion Energy Co., a subsidiary of CNNC, was registered in Shanghai with a charter capital of 15 billion yuan.
Wang Teng emphasized the advancement of China's thermonuclear energy research infrastructure: EAST achieved high-confinement mode operation for a thousand seconds, CRAFT is close to full readiness, and BEST is in the intensive assembly stage. High-temperature superconducting magnets, AI-driven plasma control, and advanced manufacturing have accelerated the transition from laboratory tests to commercial installations. The combination of government support, private capital, and technological breakthroughs allows China to potentially lead the global race in thermonuclear energy, with the target date for launching the first commercial facility around 2040.
