{"version":1,"type":"story","url":"https://digestai.news/story/alibaba-outlines-qwen-4-5-5-and-zhenwu-v900-ai-chip-for-2027","json":"https://digestai.news/story/alibaba-outlines-qwen-4-5-5-and-zhenwu-v900-ai-chip-for-2027.json","markdown":"https://digestai.news/story/alibaba-outlines-qwen-4-5-5-and-zhenwu-v900-ai-chip-for-2027.md","slug":"alibaba-outlines-qwen-4-5-5-and-zhenwu-v900-ai-chip-for-2027","headline":"Alibaba outlines Qwen 4.5, 5 and Zhenwu V900 AI chip for 2027","summary":"Alibaba detailed its AI roadmap at the Apsara Conference, focusing on its Qwen models, custom chips, and agent services. The company confirmed Qwen 4 is in training and plans Qwen 4.5 and Qwen 5, with models scaling to **5–10 trillion parameters**. It also showcased recursive self-improvement, where Qwen3.8-Max iterated **33 times** over a month, improving its score from **40 to 45** in Artificial Analysis, and refined chip designs with **10,000+ automation tool calls**, cutting chip area by **42%** without performance loss.\n\nBeyond models, Alibaba introduced the **Zhenwu V900 AI accelerator**, three times faster than its predecessor, with **216GB memory** and **1,200GB/s bandwidth**. Mass production is set for **Q1 2027**. The company also expanded its cloud infrastructure, including **AgentCore** for deploying AI agents, **Agent Security Center** for threat detection, and **OpenLake** for unified data processing, claiming **38% cost savings** and **40% faster queries**. Alibaba Cloud aims to expand its data center capacity to **20GW by 2032** to support these advancements.","keyPoints":["Qwen 4.5 and Qwen 5 models may reach **5–10 trillion parameters**, per Alibaba’s roadmap","Zhenwu V900 AI chip delivers **threefold performance** over V900 M890, launching in Q1 2027","Alibaba Cloud’s AgentCore cuts token use by **up to 67%** for tasks like customer service"],"whyItMatters":"Alibaba’s push into trillion-parameter models and high-performance chips signals a major bet on AI infrastructure. The Zhenwu V900 and cloud tools could accelerate enterprise AI adoption, while recursive self-improvement hints at future automation gains.","category":{"slug":"hardware","name":"Hardware & Compute","url":"https://digestai.news/category/hardware"},"entities":{"companies":["Alibaba","Alibaba Group","Alibaba Cloud","T-Head"],"models":["Qwen 4","Qwen 4.5","Qwen 5","Qwen3.8-Max","Qwen3.8-LiveTranslate","Qwen-Audio-3.1-TTS-Next"],"people":["Eddie Wu"]},"firstPublishedAt":"2026-09-22T07:28:00Z","updatedAt":"2026-09-23T16:44:00Z","sourceCount":2,"hasPrimarySource":false,"sources":[{"outlet":"techedt.com","title":"Alibaba lays out AI roadmap across Qwen models, chips and agent services","url":"https://techedt.com/alibaba-lays-out-ai-roadmap-across-qwen-models-chips-and-agent-services","publishedAt":"2026-09-22T07:28:00Z","type":"press","primary":false,"lead":true},{"outlet":"ibtimes.sg","title":"What Is Alibaba's Zhenwu V900 and How Does It Support the Qwen AI Roadmap?","url":"https://ibtimes.sg/what-alibabas-zhenwu-v900-how-does-it-support-qwen-ai-roadmap-94167","publishedAt":"2026-09-23T16:44:00Z","type":"press","primary":false,"lead":false}],"sourceNotes":null,"discussions":[],"thread":{"title":"Alibaba's Massive AI Infrastructure and Model Expansion","url":"https://digestai.news/thread/alibaba-unveils-zhenwu-v900-ai-chip-and-plans-over-20-gw-data-center-capacity","storyCount":3},"cite":{"text":"Digest AI, \"Alibaba outlines Qwen 4.5, 5 and Zhenwu V900 AI chip for 2027\", 22 September 2026, https://digestai.news/story/alibaba-outlines-qwen-4-5-5-and-zhenwu-v900-ai-chip-for-2027","publisher":"Digest AI","title":"Alibaba outlines Qwen 4.5, 5 and Zhenwu V900 AI chip for 2027","datePublished":"2026-09-22T07:28:00Z","url":"https://digestai.news/story/alibaba-outlines-qwen-4-5-5-and-zhenwu-v900-ai-chip-for-2027"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}