From July 17 to 20, the 2026 World Artificial Intelligence Conference (WAIC) and the High-Level Meeting on Global AI Governance were held in Shanghai.
At the conference exhibition area, the “thousand-li horses” from Beijing Haidian were particularly striking—Baidu, Moore Threads, Moonshot AI, Booster Robotics, Zhongke Huiling, Galaxy General, Zhongke WengeAI... They made a collective debut with hardcore innovations, demonstrating to the world the vigorous momentum of Haidian’s artificial intelligence.
“Fast” and “Steady” are precisely the keywords to understanding Haidian’s artificial intelligence.
Fast is the fast of “real guns and live ammunition”
Moonshot AI released its new -generation model—Kimi K3, the “world’s largest open-source model by parameter count,” with a parameter scale reaching 2.8 trillion. Booster Robotics unveiled its Booster T2, whose natural, stable and flexible locomotion performance won widespread acclaim.
From models to chips, from agents to robots, the pulse of “acceleration” beats everywhere.
Technology iteration is fast—chips are no longer showing off single units, but clusters. Moore Threads publicly demonstrated for the first time the full-stack native training results of the world’s first 5D world model “PKU EvoPhys-World,” and released three “AI Factories” based on the KUAE Intelligent Computing Cluster, supporting diverse computing power needs from training to inference, from the digital world to the physical world. Huawei Ascend 950 super node made its first real-machine appearance; Sugon brought the all-domestic 100,000-card AI super cluster Sugon 8000(Dengfeng); Tsingmicro debuted its “Reconfigurable 3.0 Technology” and the newly released RAISA software stack; Kunlunxin presented a physical super node machine, achieving full interconnect communication between cards within a single cabinet, with bandwidth increased 8-fold and inference efficiency improved 13-fold... These computing clusters are providing domestic computing power support for large model training and inference.
Intelligent evolution is fast—agents are no longer chatting with you, but working for you. As the only universal agent product selected among the conference’s “Hall of Treasures,” Baidu “Dazi” has grown at an astonishing speed, with daily average usage skyrocketing 20-fold. An increasing number of users are entrusting it with real tasks—researching materials, organizing web pages, writing reports, making PPTs, planning content, and reviewing projects.
Enterprise-level agent evolution is also advancing at rapid speed. Lenovo Lexiang 4.0 has quickly achieved enterprise-level L3 advanced intelligence. Without complex setup, a single natural language command enables Lexiang to seamlessly handle a string of tasks from product consultation, product recommendation, order placement to after-sales maintenance, and it “gets better with use” as usage frequency increases. The “Lexiang No. 1” humanoid robot, also on display on site, is the star “silicon-based employee” at Lenovo’s offline stores.
Industrial deployment is fast—large models endow robots with powerful “brain-filling” capabilities. Best World AI’s full series of “Universal World Model” products made their debut. A staff member explained: “Previously, robots mostly followed fixed programs, and changing a scene required re-debugging. Now, with the world action model, robots can ‘brain-fill’ their next few moves on their own, and flexibly handle objects they’ve never encountered before.”
Galaxy General’s heavy-duty robot Galbot S1 was stacking pallets and driving screws in the exhibition area, its power and precision effortlessly controlled; nearby, Galbot G1 was preparing breakfast, handling complex long-horizon tasks with ease, and even responding calmly to audience disturbances. These were not pre-programmed performances, but real operations under the real-time scheduling of the AstraBrain embodied large model. Zhongke WengeAI made a major release of the industry’s first decision AI product system, likening it to a “Decision AI Tree.” It roots downward into enterprise infrastructure and massive data, grows upward into cognition, reasoning and decision-making capabilities, and ultimately extends the branches of intelligence to every real business scene.
More and more Haidian AI enterprises are proving with operational systems in action and verified sets of data: this “thousand-li horse” of artificial intelligence is accelerating its gallop in Haidian.
Steady is the steady of “moistening things silently”
Only by moving steadily can one go far; only by consolidating the foundation can one innovate. The core confidence for running fast lies in running steadily.
Steady is the “reins” of the system being held with measured looseness and tightness. Currently, the national AI governance institutional system is being woven more densely at an accelerated pace, with the “Interim Measures for the Management of Generative Artificial Intelligence Services,” “Interim Measures for the Management of Artificial Intelligence Anthropomorphic Interaction Services,” and “Measures for the Identification of Artificial Intelligence-Generated Synthetic Content” being successively implemented. From algorithm filing and large model filing to content identification and scenario standardization, the institutional framework for industry development is continuously improving. At the same time, the “Qinglang • Rectifying AI Application Disorder” special action is being advanced in parallel, clearly delineating industry red lines and fortifying compliance baselines, ensuring that enterprise innovation has boundaries and development has guidelines.
Steady is also the precise delivery of service empowerment’s “post stations, provisions and runways.” If national regulation is the “reins” of measured constraint, then Haidian’s supporting services, resources, space and foundation are the fertile soil nourishing the robust growth of the industry.
On June 13, 2025, under the guidance of the Beijing Municipal Cyberspace Administration, the Haidian District Cyberspace Administration, in conjunction with Dongsheng Town, established the city’s first batch of large model ecosystem service stations at Beijing AI Origin Community—the Haidian AI Innovation Block Large Model Ecosystem Service Station. The service station directly addresses the eight core pain points of large model SMEs: “finding policies, finding funding, finding computing power, finding talent, finding partners, finding intelligence, finding space, and finding scenarios.”
In the past, enterprises had to “cross the city” to run procedures to learn about filing policies and compliance requirements. Today, filing guidance has been brought down to the ecosystem service stations. The service stations regularly carry out “one-stop” public welfare services including policy publicity, resource matchmaking, and technical exchanges. As of July 2026, a cumulative total of 141 generative AI services have completed filing, making it the first prefecture-level region in the country to achieve the “hundred-model scale.”
To let the “thousand-li horses” gallop better, Haidian has prepared ample “provisions.” According to the “Several Measures for Zhongguancun Science City to Accelerate the Construction of an AI Industry Highland with Global Influence,” more than RMB 1 billion is planned annually for breakthrough R&D investment, computing power subsidies, data incentives, scenario support and other funds.
In terms of space, taking “dual ten, dual weighting” as the lever, Haidian is building the Century-old Jingzhang AI Innovation Belt and constructing a “two zones, one belt” ecosystem: the southern AI Origin Community strengthens technology sourcing, the northern Zhongguancun AI Beowei Community focuses on advanced incubation, and the Jingzhang Relics Park Innovation Belt connects north and south.
And underpinning all of this is Haidian’s profound innovation foundation. Dense clusters of universities and research institutes, massive AI enterprise clusters, abundant computing power and data resources, and diverse scenario deployment spaces together constitute a complete closed loop from technology breakthrough to industrial transformation.
