AI
Alibaba Plans a 5 to 10 Trillion Parameter Model, and the Zhenwu V900 Chip Is the Bigger Story
The parameter number grabs the headlines. At its annual Apsara Conference in Hangzhou on Tuesday, Alibaba CEO Eddie Wu said the company plans to train a Qwen model with 5 to 10 trillion parameters, a leap past the current flagship Qwen 3.8 Max at 2.4 trillion. The numbers are enormous, and they point at something bigger than one model.
The bigger story is the silicon underneath it. Alibaba also unveiled the Zhenwu V900, a new AI accelerator from its T-Head chip division, and called it the most powerful AI chip in China. The V900 delivers three times the performance of the Zhenwu M890 released in May, carries 216 gigabytes of memory with 1,200 gigabytes per second of chip to chip bandwidth, and supports FP8 and FP4 natively, the low precision formats that make inference faster and cheaper.
Scale is the point. The V900 slots into an upgraded supernode server that scales to clusters of up to 500,000 cards, and Alibaba plans mass production and commercial release in the first quarter of 2027. The company says it will bring its AI supernodes online at commercial scale this quarter, and its existing Zhenwu chips already serve more than 650 customers across automotive, finance, energy, and manufacturing.
Then comes the land grab for power and floor space. Wu set a target of more than 20 gigawatts of global data center capacity for Alibaba Cloud by 2032, backed by a commitment of more than 53 billion dollars over three years, with a 10.2 billion dollar follow on share offering in Hong Kong in August helping fund the push. Demand is outpacing supply, and Wu said the industry's mid to long term demand far outpaces its supply capabilities.
The frame Wu put around it all is worth hearing. He said machine intelligence currently represents less than 3 percent of human capacity, leaving vast room to grow, and called AI coding the light bulb of the machine intelligence era. Investors heard it. Alibaba shares climbed about 3 percent in Hong Kong on Tuesday, while Tencent surged more than 7 percent on the same AI optimism that lifted global chip stocks.
The model roadmap runs in parallel. Qwen 4 is in training now, with Qwen 4.5 and Qwen 5 on the roadmap, and Alibaba says its researchers are making progress on models that identify weaknesses, run experiments, and produce training data with less human involvement. The company frames all of it as a march toward increasingly autonomous systems, with the silicon, the data centers, and the models advancing as one program. Wu put it in historical terms, saying today's AI development sits at the early stages of electrification, and AI coding is the light bulb of the machine intelligence era.
For you, the takeaway is that the AI race is becoming a full stack contest. Whoever controls the models, the chips, and the gigawatts controls the economics of intelligence for the next decade. Alibaba just showed it intends to own all three layers at once.
Quick answers
What is this story about?
The parameter number grabs the headlines. At its annual Apsara Conference in Hangzhou on Tuesday, Alibaba CEO Eddie Wu said the company plans to train a Qwen model with 5 to 10 trillion parameters, a leap past the current flagship Qwen 3.8 Max at 2.4 trillion. The numbers are enormous, and they point at something bigger than one model.
Why does this story matter?
For you, the takeaway is that the AI race is becoming a full stack contest. Whoever controls the models, the chips, and the gigawatts controls the economics of intelligence for the next decade. Alibaba just showed it intends to own all three layers at once.
Sources
- Reuters: Alibaba plans 5 to 10 trillion parameter model, unveils Zhenwu V900
- CoinCentral: Alibaba stock jumps on new AI chip and data center plans
- TechWire Asia: Alibaba AI chip and 20GW plan meet a supply squeeze
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.