中文翻译
摘要: Alibaba Launches Zhenwu M890 Chip, Claiming 3x Performance Leap to Challenge Nvidia
Alibaba unveiled its next-generation AI chip, the Zhenwu M890, claiming a threefold performance increase over its p...
正文
Alibaba Launches Zhenwu M890 Chip, Claiming 3x Performance Leap to Challenge Nvidia
Alibaba unveiled its next-generation AI chip, the Zhenwu M890, claiming a threefold performance increase over its predecessor and targeting the rapidly growing market for AI agents. Developed by subsidiary T-Head, the chip features 144 GB of GPU memory and will be available immediately through a new 128-accelerator server system. The launch, strategically timed ahead of Nvidia's quarterly earnings, highlights China's accelerating push for semiconductor self-sufficiency amid tightening U.S. export controls. Alibaba also outlined a multi-year roadmap with successive chips planned through 2028 and introduced the Qwen 3.7-Max AI model. While analysts note the technology still trails Western leaders, Alibaba's $53 billion infrastructure commitment and 560,000 units already shipped signal serious competitive intent in a market increasingly dominated by domestic players like Huawei.
Key Elements
Zhenwu M890
Qwen 3.7-Max
SemiAnalysis
Donald Trump
Howard Lutnick
Cambricon
Alibaba Group Holding Ltd. (BABA) introduced its next-generation artificial intelligence processor, the Zhenwu M890, at its annual cloud summit in Hangzhou on Wednesday, marking a significant escalation in China's push for semiconductor self-sufficiency. The chip, developed by Alibaba's semiconductor design subsidiary T-Head, is purpose-built for AI "agents"—autonomous systems capable of executing complex, multi-step tasks with minimal human intervention. Alibaba claims the M890 delivers three times the performance of its predecessor, the Zhenwu 810E, and is engineered to handle the massive memory and communication demands of agentic AI workloads, where models must retain long contextual threads and coordinate in real time.
The technical specifications reveal a processor configured with 144 GB of GPU memory and an interchip bandwidth of 800 GB per second. T-Head stated it has alread
采集时间: 2026-05-20 20:16:47
AIAPIAI人工智能技术