Alibaba is preparing Qwen AI models with as many as 10 trillion parameters, launching a new in-house AI chip and targeting more than 20GW of global data centre capacity by 2032 as it expands across the full artificial intelligence technology stack.
The plans, announced at Alibaba Cloud's annual Apsara conference in Hangzhou, span foundation models, semiconductors, AI computing clusters and data centre infrastructure.
Alibaba CEO Eddie Wu said demand for AI computing is already outpacing the company's ability to add capacity, with Alibaba Cloud preparing to bring AI supernodes online commercially this quarter.
Qwen models could scale to 10 trillion parameters
Alibaba's Qwen team plans to develop models containing between 5 trillion and 10 trillion parameters as the company targets increasingly complex and longer-horizon AI tasks.
Alibaba's current flagship, Qwen 3.8 Max, contains 2.4 trillion parameters.
The company said its next-generation Qwen 4 model is currently being trained, while future Qwen 4.5 and Qwen 5 models are expected to scale into the 5 trillion to 10 trillion parameter range.
Wu linked the development programme to Alibaba's pursuit of artificial superintelligence, referring to AI systems capable of surpassing human capabilities.
He also said the Qwen team had made "meaningful" progress toward models capable of improving themselves by identifying weaknesses, conducting experiments and generating training data with limited human involvement.
The company has not provided training-compute requirements, release dates or performance benchmarks for the planned larger models.
Zhenwu V900 AI chip targets massive compute clusters
Alibaba also unveiled the Zhenwu V900, a new AI processor developed by its T-Head semiconductor unit.
According to Wu, the V900 delivers three times the performance of Alibaba's previous-generation M890 chip, which was launched in May.
The new processor is designed to operate in clusters containing as many as 500,000 chips, creating large-scale computing systems capable of training and running increasingly demanding AI models.
Alibaba plans to begin mass production and commercial availability of the V900 during the first quarter of 2027.
Wu said the company expects significant growth in annual AI chip shipments, although Alibaba did not disclose current shipment volumes, manufacturing partners or detailed technical specifications for the V900.
Alibaba Cloud targets more than 20GW of data centre capacity
The infrastructure ambitions extend beyond individual processors.
Wu set a target for Alibaba Cloud's global data centre capacity to exceed 20GW by 2032, illustrating the scale of computing and power infrastructure the company expects will be required as AI workloads grow.
Alibaba Cloud is also preparing to bring AI supernodes online at commercial scale during the current quarter.
Wu described customer demand for AI as "exceptionally robust" and said it is accelerating Alibaba Cloud's revenue growth.
Supply constraints, however, remain a significant challenge.
"The industry's mid-to-long-term demand far outpaces our supply capabilities," Wu said.
The company did not provide a current data centre capacity figure alongside its 20GW target, making it difficult to quantify the infrastructure expansion required to reach the 2032 goal.
Alibaba builds across the AI technology stack
The announcements show Alibaba attempting to control several increasingly interconnected layers of AI infrastructure.
Qwen provides the foundation-model layer, T-Head is developing the processors required for AI computation, Alibaba Cloud supplies computing services and supernodes, and the company's data centre expansion provides the physical infrastructure needed to support those workloads.
That vertical approach could become increasingly important as AI models require larger amounts of compute and cloud providers face constraints around processors, power and data centre capacity.
Wu argued that the most consequential applications of machine intelligence have yet to emerge, comparing the current development of AI with the early stages of the Industrial Revolution.
He predicted that machines could eventually generate more than 1,000 times the combined "thinking" of humanity, compared with less than 3% today. The statement represents Wu's long-term projection rather than a measurable industry forecast.
Source: MEA Tech Watch Press Reporter
