← China Tech Daily
About Editorial Standards Corrections Privacy Policy
Subscribe →
China Tech · Daily chinatechdaily.org
JULY 23, 2026
Your morning brief on Chinese technology. Published by the AI Policy Institute
Bottom Line Up Front

Alibaba ran its 2.4-trillion-parameter Qwen3.8 model on its own T-Head chips, saying its Lingjun Zhenwu M890 super-node is the first in China to run a model above 2 trillion parameters. China began pushing its new AI agent interoperability standards into industry this week, issuing the first 2,000-plus agent identity codes and releasing an English translation for international outreach. The measures aim to make agents from rival Chinese vendors interoperable by default. MOFA spokesperson Lin Jian rejected U.S. criticism of Chinese AI firms at a July 23 briefing, calling such practices politicization and weaponization of science and technology.

ITOP STORIES
 
Alibaba runs 2.4-trillion-parameter Qwen model on its own AI chips for the first time
Alibaba Cloud says its super-node is the first in China to run a model of more than 2 trillion parameters. A super-node is a rack-scale system of tightly interconnected AI chips that software treats as a single large accelerator. The Lingjun Zhenwu M890 super-node links 64 Zhenwu M890 chips from Alibaba's in-house design arm T-Head at 800GB/s, with 9TB of memory. Alibaba Cloud's Bailian model-serving platform now runs Qwen3.8, a mixture-of-experts design whose 2.4 trillion parameters activate only a subset per query. The setup delivers up to 1.5 times inference performance in agentic scenarios, and Alibaba has said Qwen3.8 may be the most powerful model apart from Anthropic's Fable 5.
Read at IT Home ↗·Read at IT Home ↗
China releases national standards for AI agent identity codes and cross-system connection
China is setting national standards for how AI agents identify themselves and connect across systems, positioning Beijing to write the global rulebook for agent interoperability. The State Administration for Market Regulation (SAMR), China's principal market regulator, approved the seven GB/Z185-2026 guidance-type standards on June 26. Officials held a promotion conference in Beijing's Zhongguancun tech district this week to push adoption. The framework spans overall architecture, agent identity codes, identity management, capability description, cross-domain discovery, collaborative interaction and tool invocation, with Volcano Engine, Xiaomi, Kuaishou and Lenovo among the drafters. More than 2,000 agent identity codes have already been issued to industry participants. An English translation was released at the 2026 World Artificial Intelligence Conference International Standardization Forum, with Russian and Arabic versions planned for outreach to the Association of Southeast Asian Nations (ASEAN).
Read at IT Home ↗
China's Foreign Ministry rejects U.S. criticism of Chinese AI firms as politicization
China's Foreign Ministry rebutted U.S. officials' criticism of Chinese AI, framing such practices as politicizing and instrumentalizing science, technology and trade. Ministry of Foreign Affairs (MOFA) spokesperson Lin Jian said at a July 23 press briefing that China consistently opposes politicizing and instrumentalizing science, technology, and economic and trade issues, and that such conduct only disrupts global AI development and serves no party's interests. Lin was responding to a question about a White House official's accusation that a Chinese AI startup stole Anthropic's most advanced large language model.
Read at Yicai ↗·Read at China News Service ↗
Shanghai opens STAR Market to fusion, embodied AI, quantum and brain-computer interface firms
Shanghai is opening the STAR Market, the Shanghai Stock Exchange's science and technology innovation board, to channel patient capital into frontier hard-tech firms that are not yet profitable. Municipal authorities released a 20-measure package on July 23 that expands the market's "fifth listing standard," its pathway for admitting companies on technology potential rather than current profit. Target sectors include controlled nuclear fusion, quantum computing, brain-computer interfaces, large language models and embodied AI, which pairs an AI system with a physical body such as arms, grippers or sensors. Shanghai will also accelerate a Social Security Sci-Tech Innovation Fund and issue listing review guidelines for AI and low-altitude economy firms, meaning commercial drone and short-range aircraft operators, under the fifth-standard pathway.
Read at Wallstreetcn ↗
IIPOLICY & REGULATION
 
China's Commerce Ministry names humanoid robot R&D as a target for foreign investment
China's Commerce Ministry is naming humanoid robot research as a sector where Beijing wants foreign capital, adding an inbound investment lever atop the humanoid robot industrial policy push. Meng Huating, director-general of foreign investment administration at the Ministry of Commerce (MOFCOM), spoke at a July 23 press briefing. Advanced manufacturing targets include organic polymer materials and high-efficiency magnetic-levitation power equipment, while modern services targets include humanoid robot R&D and high-end shipping services. The Ministry plans to improve foreign R&D center support policies and ease entry procedures for senior foreign talent.
Read at Yicaistate media ↗
Beijing turns AI agent stack into its own industrial policy category with token and agent service vouchers
Beijing has made AI agents a stand-alone industrial policy category, using subsidy vouchers to steer compute, tokens and agent services toward domestic providers. The Beijing Municipal Development and Reform Commission (Beijing NDRC), the city's economic planner, issued the Measures on Accelerating Agent-Led Development on July 23 with partner departments. Token-as-a-Service, Agent-as-a-Service and Result-as-a-Service are named as new categories under the city's small-business service voucher program, with districts encouraged to add token and agent vouchers atop existing compute vouchers. Beijing added 22,000 petaflops of intelligent computing capacity, a measure of AI compute throughput, in the first half of 2026, lifting supply to 82,000 petaflops and targeting 130,000 by year-end.
Read at Yicaistate media ↗·Read at South China Morning Post ↗
IIIROBOTICS & AUTONOMOUS SYSTEMS
 
Hong Kong clears Baidu's Apollo Go for fully driverless robotaxi testing on Airport Island
Hong Kong has cleared what the Transport Department calls the world's first fully driverless public-road testing in a right-hand-drive, left-traffic system, meaning steering on the right and traffic on the left. The Hong Kong Transport Department issued the city's first driverless testing license on July 23 to Apollo Go, Baidu's autonomous ride-hailing brand. Fully driverless testing without an in-vehicle safety operator begins July 27 on Airport Island. Since February 2026, testing on the same stretch has added a remote-control-center operator alongside the in-vehicle backup operator, and the department said it found no abnormal conditions after reviewing test reports and conducting site inspections. Apollo Go has said the license is its first in a right-hand-drive, left-traffic jurisdiction.
Read at IT Home ↗
IVQUANTUM & FRONTIER SCIENCE
 
China moves to set up national standards body for space-based solar power
China is drafting the rulebook for space-based solar power before the sector has commercial deployments, a familiar pattern of standards-first industrial policy. The Solar Photovoltaic System Application Standardization Technical Committee (CSTM/FC03/TC22), which sits under the building materials standardization field committee of the China Standards for Materials and Testing, is preparing to establish a Space Photovoltaic Working Group. The group's scope covers surveying domestic and international standards, building a standards framework, and conducting research on extreme-environment testing, reliability evaluation, on-orbit verification and system integration for solar power generated in orbit rather than on Earth's surface.
Read at Eastmoney ↗

Thanks for reading. See you tomorrow.— Daniel

Comments or corrections? hello@theaipi.org

 
Editor-in-Chief:Daniel Colson
Managing Editor:Min Goodman-Cheng
Lead Software Engineer:Christopher Käck
Editorial Advisor:Carrie Adams
Research Manager:Philip Wieczorek
Operations Specialist:Joanne Chua
China Policy Fellow:Tyler Liggett
Research Assistants:Claude Code + Codex
 

View in browser · LinkedIn

AI Policy Institute · 700 Pennsylvania Avenue SE, Washington, DC 20003

China TechDaily
About Editorial Standards Corrections Privacy Policy
© 2026 AI POLICY INSTITUTE · WASHINGTON D.C.