Ce sa întâmplat
At its annual Apsara Conference in Hangzhou, Alibaba Group announced a strategic roadmap to advance its artificial intelligence capabilities, focusing on hardware, model development, and cloud infrastructure. The company introduced the Zhenwu V900, which it claims is the most powerful AI chip currently developed in China. Additionally, Alibaba’s Qwen development team revealed plans to scale its future Qwen 5 model to between 5 trillion and 10 trillion parameters, aiming to achieve artificial superintelligence (ASI).
Alibaba Group chairman Joe Tsai and CEO Eddie Wu Yongming detailed a 'full-stack' AI strategy at the Apsara Conference, emphasizing the integration of AI models, chips, and cloud services. The company aims to scale its global data-center capacity to over 20 gigawatts by 2032 to support these initiatives.
The newly unveiled Zhenwu V900 AI chip is reported to offer three times the performance of its predecessor, the M890. According to T-Head, Alibaba's semiconductor subsidiary, the chip features 216 gigabytes of high-bandwidth memory and 1.2 terabytes per second of inter-chip bandwidth. Mass production is slated for Q1 2027.
The Qwen development team, led by Liu Dayiheng, outlined a roadmap for the Qwen 4, 4.5, and 5 series. The Qwen 5 model is intended to reach 5 to 10 trillion parameters. For context, the current flagship, Qwen 3.8-Max, utilizes 2.4 trillion parameters.
Alibaba also launched 'Qwen Intelligence,' an agent platform designed for smartphone integration, and confirmed ongoing collaboration with Apple regarding AI features for iPhones in the Chinese market.
De ce contează
This announcement represents a significant escalation in the domestic competition for AI dominance in China, particularly as firms navigate U.S. export restrictions on advanced semiconductors. By developing proprietary hardware like the Zhenwu V900 and scaling model parameters to the 10-trillion range, Alibaba is attempting to reduce reliance on foreign silicon while positioning itself to compete with global frontier model developers. The integration of these chips into a massive cloud infrastructure, targeting 20 gigawatts of capacity by 2032, underscores the company's strategy to commoditize intelligence at scale.
The development of the Zhenwu V900 is a direct response to the challenges posed by U.S. export controls, which have limited access to high-end Western AI chips. By building proprietary hardware, Alibaba seeks to ensure its cloud infrastructure remains viable for training frontier models.
The push toward 10-trillion- models reflects a broader industry trend toward 'agentic AI,' where models are expected to execute complex, multi-step tasks across various applications over extended periods, rather than simply generating text or images.
Alibaba's commitment to a 20-gigawatt data center capacity by 2032 highlights the massive energy and infrastructure requirements necessary to sustain the next generation of machine intelligence, positioning the company as a primary provider of AI-as-a-commodity.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').What is the best response when AI Models Explained makes a mistake in production?
Ce să urmărești în continuare
Market observers should monitor the mass production timeline for the Zhenwu V900, which is scheduled for the first quarter of 2027. Furthermore, the actual performance of the proposed Qwen 5 model remains to be seen, as the company has only provided a roadmap for its development. The effectiveness of these domestic chips in training large-scale models compared to the Nvidia processors currently used by many Chinese developers will be a critical indicator of the success of China's semiconductor self-sufficiency efforts.
The transition from the current reliance on Nvidia processors to the proprietary Zhenwu architecture will be a key metric for Alibaba's long-term independence in the AI sector.
Future updates on the Yitian 730 CPU, expected in 2027, will indicate how well Alibaba can optimize its in-house microarchitecture for agentic AI tasks.
Competition with Huawei's Ascend 960-powered Atlas 960 SuperPoD will likely intensify, as both companies vie for dominance in the domestic Chinese AI computing market.