日本語
2026-09-22 Evening edition
Evening editionAI News Daily
2026-09-22
Alibaba unveiled a new AI chip, the Zhenwu V900, alongside plans to expand data center capacity to over 20GW. Hugging Face's Transformers library now supports llama.cpp quantized formats.
Today's Highlights
Company NewsAlibaba Unveils New AI Chip "Zhenwu V900," Plans 20GW Data Center Expansion
2026-09-22 15:50
Feature UpdateHugging Face's Transformers Adds Support for llama.cpp Quantized Formats
2026-09-22 09:00
Alibaba Unveils New AI Chip "Zhenwu V900"
Announced: 2026-09-22 15:50
- Triples the performance of the prior-generation "Zhenwu M890" and supports clusters of up to 500,000 units
- Mass production is planned for Q1 2027
Why it matters: Shows China accelerating performance gains and mass-production plans for domestically made AI chips even under U.S. export controls.
KEY CONTEXTAnnounced at the Apsara Conference in Hangzhou. Alibaba shares rose in both Hong Kong and U.S. trading following the announcement.
Alibaba to Expand Data Center Capacity Beyond 20GW by 2032
Announced: 2026-09-22 15:50
- Long-term data center expansion plan unveiled alongside the new chip announcement
- Target horizon extends to 2032
- Positioned to support the localization of China's AI computing infrastructure
20GW
Why it matters: A scale of investment worth tracking for its implications on the global balance of semiconductor and data center spending.
Hugging Face's Transformers Adds Support for llama.cpp Quantized Formats
Announced: 2026-09-22 09:00
- The widely used Transformers library can now directly handle llama.cpp-format quantized models (e.g., GGUF)
- Announced on Hugging Face's official blog
- Lets open-source LLMs be used in lightweight quantized form directly through the standard library
Why it matters: Broadens the options for model selection and cost optimization when deploying AI.
Trend Overview
- China (via Alibaba) is accelerating performance gains and mass-production plans for domestic AI chips, advancing localization of its computing infrastructure
- Major AI providers continue steady ecosystem-level library enhancements, such as support for lightweight formats
- The 9/22 morning-to-evening window was dominated by previously reported news; items arising after the deadline (e.g., the AI-regulation speech at the UN) carry over to the next edition
Summary
Key Takeaways
- Alibaba announced a new AI chip with 3x performance and mass-production plans
- Plans to expand data center capacity beyond 20GW by 2032
- Hugging Face's Transformers now supports quantized formats
What to Watch Next
- Start of Zhenwu V900 mass production (planned for Q1 2027)
- Progress on data center expansion
- Follow-up on items carried to the next edition (e.g., the AI-regulation speech)