In the past 24 hours (September 25th), my country's Ministry of Industry and Information Technology announced that over 1,500 large-scale AI models have been released, with simultaneous updates to operational and computing system data. Alibaba launched Qwen3-Max (over 1 trillion parameters) at the Yunqi Conference and announced global data center expansion and collaboration with Physical AI. Industry conferences and trade shows have signaled the widespread adoption of AI. Internationally, Google's "Search Live" service has officially launched in the US, OpenAI announced Stargate expansion targets and investment updates, and Microsoft integrated the Anthropic model into 365 Copilot. Meanwhile, AI computing hosting and financing are accelerating.
1. Ministry of Industry and Information Technology: my country has released more than 1,500 large-scale artificial intelligence models
- Disclosure on September 25: More than 1,500 large models have been released in China; the number of 5G mobile phone users exceeds 1.1 billion, and 91% of telecommunications services are handled online (based on the opening of the China International Information and Communications Exhibition).
- 5G has covered 86 major industries in the national economy, the "computing power Internet" has been launched, and industry and infrastructure have advanced in tandem.
- The above data show that the big model is "going deeper and more practical", laying a solid foundation for "artificial intelligence +" actions and industrial applications.
2. Alibaba Cloud releases Qwen3-Max (>1 trillion parameters), announces global expansion and collaboration with Physical AI
- September 24th Yunqi Conference: Qwen3-Max has a parameter scale of over 1 trillion, emphasizes code and autonomous agent capabilities, and claims to be superior to some competitors in indicators such as Tau2-Bench.
- Announced cooperation with NVIDIA in the field of Physical AI, and planned to invest RMB 38 billion (cumulatively) in AI infrastructure over three years, and stated that further additions will be made.
- The simultaneous launch of the multimodal Qwen3-Omni and the planning of new data centers in multiple overseas locations demonstrate the three-pronged approach of "model + computing power + going global".
3. JD.com's "Enjoy AI" launches AI panorama and new products
- JDDiscovery Conference on September 25: Release of JD AI Panorama, covering intelligent computing, models, platforms and scenarios.
- Launch "Jingxi", the next-generation shopping and life service portal, as well as "He, She, It" digital human assistant and JoyInside and other characters.
- Gradually open up high-quality industry model data sets to large models and embodied intelligence enterprises, emphasizing three years of continuous investment and ecological driving effects.
4. AI hardware became a highlight at the Digital Trade Fair, with domestic AI devices attracting attention.
- The 4th Global Digital Trade Expo opened in Hangzhou on September 25, showcasing AI hardware and applications.
- Alibaba DingTalk's first AI hardware, DingTalk A1, has gone viral, attracting foreign businesspeople to experience it, demonstrating the trend of integrating enterprise and personal scenarios.
5. China Unicom releases progress on "Digital Intelligence Empowerment, AI Connects the Future"
- Disclosure on September 25: A series of activities including "Hundreds of Links, Thousands of Communications, and Tens of Thousands of (Anhui) Enterprises' Intelligence" were launched during the World Manufacturing Conference.
- Demonstrate AI solutions for government, enterprises and industries, emphasizing the synergy of "connection + computing power + application".
6. Google Search Live is now available in the US: real-time voice conversation and camera sharing
September 24-25: Search Live officially launched for US English users, without the need for a Labs application. A new "Live" entry was added to the app.
- Supports real-time voice conversation in AI Mode, and can share mobile phone camera images for scene understanding and guidance.
- At the same time, the Gemini Live API introduced a preview version of the native audio model to improve the naturalness and stability of the voice agent.
7. Microsoft integrates the Anthropic model into Microsoft 365 Copilot
September 24: Switching between Claude Sonnet 4 and Claude Opus 4.1 is now available in Copilot Researcher and Copilot Studio.
- This move will maintain the dominant position of OpenAI models while improving the selectivity and task matching of multiple models.
- A “combination punch” that reflects the ecological trend of “multi-model parallelism” and enterprise-level AI procurement.
8. OpenAI, Oracle, and SoftBank Expand Stargate: Targeting $500 Billion/10GW
September 23-25: Announced five new US AI data center sites, bringing Stargate's planned capacity to nearly 7GW and total investment over three years to over $400 billion.
- The path to achieving the $500 billion/10GW commitment by the end of 2025 is clear, with the creation of tens of thousands of jobs.
- Collaborate with OCI and energy to provide a solid computing power base for next-generation models and reasoning.
9. Cipher Mining signs a $10 billion option for hashrate hosting, with an initial contract value of $3 billion.
September 25: Cipher and Fluidstack sign a 10-year HPC/AI hosting agreement, giving Google a 5.4% equity interest (warrants).
- The initial contract value is approximately US$3 billion, with the goal of delivering approximately 168MW of critical IT load in Texas by September 2026.
- Mining companies are accelerating their path to transforming HPC and AI infrastructure, with capital expenditures and long-term electricity lock-in going hand in hand.
10. The Washington Post Global AI Summit: IMF warns that "60% of jobs in developed economies will be affected"
September 24: The IMF President estimated that AI could impact employment in developed economies by 60% and globally by 40%, and warned of the risk of financial instability from “overinvestment” in AI.
- The guests focused on AI governance, computing power and data accessibility in the Global South, and the response of the education system.
- The discussion emphasized "inclusive innovation" and cross-sectoral governance, and warned against the "AI divide."
Frequently Asked Questions (Q&A)
Q: What does “1500+” in my country’s large model mean?
A: As of September 25, more than 1,500 large models have been released in China, with over 1.1 billion 5G users, 91% of which are processed online, covering 86 major industries. This shows that the supply of models and network/computing infrastructure have matured in coordination, and industrial applications have entered an accelerated period from "usable to easy to use."
Q: What is the core attraction of Qwen3-Max?
A: The parameter scale exceeds 1 trillion, emphasizing code and autonomous agent capabilities, and leading in indicators such as Tau2-Bench; it is supported by Alibaba's three-pronged advancement of "model + computing power + globalization" (including Physical AI cooperation with NVIDIA and multi-national data center planning).
Q: How is Google Search Live different from regular voice search?
A: It supports "real-time conversation + camera sharing", that is, you can talk and watch at the same time, and AI can understand the camera image and give operation instructions or search paths; it is currently popular among English users in the United States, and the entrance is "Live" in the Google App.
Q: What changes will Microsoft bring to Copilot by bringing Anthropic?
A: Companies can freely switch between Claude and OpenAI models by task in Researcher and Copilot Studio, improving performance/options in complex searches, long-form thinking, or compliance scenarios, in line with multi-model strategies and supply chain decentralization trends.
