Alibaba

Alibaba Group unveiled new artificial intelligence chip technologies and ambitious model plans at its annual conference in Hangzhou, including what CEO Eddie Wu described as China's most powerful AI chip. The company also outlined plans to train a far larger AI model and expand its cloud computing infrastructure amid surging demand. Wu said the new Zhenwu V900 chip delivers three times the performance of its predecessor Zhenwu M890, and Alibaba plans to train a model at 5 to 10 trillion parameters, up from 2.4 trillion in its Qwen3.8-Max. Alibaba expects to exceed 20 gigawatts of computing capacity by 2032, though global supply chain shortages are slowing scaling, while Chinese-designed chips are gaining ground as US export curbs limit Nvidia sales to China.

At Alibaba Cloud's annual conference in Hangzhou on September 21-22, 2026, CEO Eddie Wu announced T-Head's new Zhenwu V900 AI chip, which Alibaba said delivers three times the performance of its May-released predecessor (Zhenwu M890). The Zhenwu V900 supports both high-precision model training and ultra-low-precision i The dual announcement pairs next-generation silicon with a substantially larger planned model, signaling Alibaba's vertical-integration strategy across the AI stack. Alibaba's Hong Kong-listed shares rose 3.1% on the news, outperforming the Hang Seng Tech Index. The piece frames competition with DeepSeek, Moonshot AI,

The Star's syndication of the Bloomberg report (dated Aug 27, 2026) confirms that Alibaba released Qwen3.8-Flash on August 26 with downloadable weights and 125 billion parameters, and adds architectural context: the release uses an architecture designed for the next generation of models in the Qwen 4 series, which Alib This candidate is technically useful because it independently names the Qwen 4 series architecture underlying Qwen3.8-Flash and quantifies Alibaba's three-year AI spend at more than 380 billion yuan (~US$57 billion), with a US$10.2 billion Hong Kong follow-on raise that week. It also contextualizes Qwen3.8-Flash alongs

On August 26, 2026, Alibaba opened the international edition of QwenWork, its workplace AI agent platform, in public beta for users outside China. The web and desktop product is initially available in English and simplified Chinese, with additional languages planned, and targets users in Asia, the Middle East, and Lati Users can ask the system to research topics, work with local files, prepare office documents, generate and host a web application, or complete browser-based tasks, with support for text, images, audio, and video and a scheduling function for recurring work. QwenWork introduces an opt-in personalization feature called A

Alibaba's Qwen team announced plans to open-source Qwen3.8-Flash-Next alongside an FP8 variant at 11 p.m. Beijing time on August 26, 2026, according to TechNode. The model is described as a multimodal mixture-of-experts system built on the next-generation Qwen4 architecture. Qwen indicated the early release is intended The move signals a staged rollout of the Qwen4 generation, with Qwen3.8-Flash-Next serving as a preview/exploration vehicle ahead of flagship family releases. An FP8 quantization variant is being released concurrently, which is relevant for inference cost and hardware efficiency. Because the candidate itself notes that

On August 26, 2026, Alibaba released the latest model in its Qwen series, dubbed Qwen3.8-Flash, making it downloadable and revealing its open weights. According to Alibaba's statement, the model has 125 billion parameters and is positioned as a lower-priced offering intended to drive global adoption of its flagship AI. The Qwen3.8-Flash launch is part of Alibaba's intensified AI investment cycle; the company has pledged more than 380 billion yuan (~US$57 billion) over three years toward AI development and infrastructure, and raised roughly US$10.2 billion in a Hong Kong follow-on offering that week to fund the effort. Although the Bl

A technology.org writeup (August 3, 2026) adds context to the Qwen3.8-Max launch, confirming the 2.4-trillion-parameter figure and the 95-billion active-parameter MoE design, and situating it just behind Moonshot AI's Kimi K3 at 2.8 trillion parameters. The article notes Qwen3.8-Max became the top-ranked Chinese text m The piece frames the parameter-count arms race among Chinese providers as a deliberate developer-marketing play, since most Chinese flagship releases are open-weight while OpenAI, Anthropic, and Google disclose nothing about closed-model size. It also reiterates that raw parameter totals don't guarantee better quality,

Alibaba released Qwen3.8-Max on August 3, 2026, its largest Qwen-family model to date, with 2.4 trillion total parameters in a mixture-of-experts design that activates 95 billion at a time and supports a 1-million-token context window. The model is immediately available via API through QwenCloud, and Alibaba confirmed The release emphasizes long-horizon agentic capabilities: Alibaba reports Qwen3.8-Max ran fully autonomously for more than 16 days in a self-built software repository, accumulating 265 commits, 127 pull requests, and 151 issues, and in a separate timed competition finished ahead of 87% of 526 human teams. Multimodal us