The most powerful AI chip in China today, delivering three times the performance of its predecessor, the Zhenwu M890.
That is how Eddie Wu described the processor Alibaba unveiled on Tuesday. The chief executive was speaking at Apsara, the company’s annual cloud conference, in Hangzhou.
The chip is the Zhenwu V900. Alibaba said in a post on X that it comes from T-Head, its chip design unit. The company says it handles both high precision model training and ultra low precision inference.
Two other announcements came with it. Alibaba plans models of up to 10 trillion parameters, and 20GW of data centre capacity by 2032.
What is inside the V900
T-Head has published a basic spec sheet for the chip. Simon Sharwood reported the numbers for The Register. The sheet lists 216GB of memory and 1,200 GB/s of inter-chip interconnect bandwidth.
The processor supports FP32 down to FP4 natively. T-Head reworked the Tensor Core arithmetic unit for better instruction precision in FP8 and FP4. The chip also carries more scaling factor formats and block size configurations under MXFP8 and MXFP4.
Wu said a single cluster can hold up to 500,000 of the chips. The resulting machine will power frontier model training and inference, he said. Alibaba launched the M890 in May.
Mass production and commercial release are set for the first quarter of 2027. Liam Mo and Eduardo Baptista reported for Reuters that Wu expects significant growth in annual AI chip shipments.
Alibaba will refresh the lineup every year, Bloomberg reported. It put deliveries of the Zhenwu family at 560,000 units to more than 400 external customers. Alibaba’s own post puts the figure higher, at more than 650 customers. They span industries from cars to language models and embodied intelligence.
The company also set out a roadmap for its next generation Yitian server processors. They are built for agentic AI work and due in 2027.
Qwen 5 is planned at 10 trillion parameters
Alibaba is already training Qwen 4. Wu named the two models after it, Qwen 4.5 and Qwen 5. Both should scale to between five and 10 trillion parameters.
Qwen 3.8 Max, the current flagship, has 2.4 trillion. Alibaba called it its most capable model in August.
Wu framed the jump as a route to artificial superintelligence. The larger models target longer and more complex tasks.
He also said the Qwen team is working on recursive self improvement. A model identifies its own weaknesses, runs experiments and generates training data with little human involvement. He described the progress as meaningful.
Twenty gigawatts by 2032
The third announcement was infrastructure. Wu set a target for Alibaba Cloud to pass 20GW of global data centre capacity by 2032.
Citigroup estimates that much capacity could drive more than $160bn of external revenue for the cloud division, according to Bloomberg.
The commercial property firm Cushman and Wakefield counts 37.7GW of data centres under construction in the United States alone. SpaceX had roughly 1.4GW of AI computing capacity in the middle of this year. It wants to pass 10GW in 2027, the Associated Press reported.
Supply is the constraint Wu named. He said global shortages across the AI data centre supply chain are limiting how fast Alibaba can scale its compute, and that mid to long term industry demand far outpaces what the company can supply.
Alibaba Cloud will start bringing its AI supernodes online at commercial scale this quarter. Wu did not say where the new data centres will be built.
What the push costs
Alibaba has committed more than $53bn over three years to its AI expansion, and raised about $10.2bn in a follow-on share sale in August.
The spending shows in the accounts. Quarterly profit fell by 75% after a spree of almost $10bn.
Executives expect annualised revenue from AI products to reach $10bn this quarter, and say the company can recoup its investment within three years. The target for cloud and AI revenue is $100bn in five years.
Investors took the announcement well. Alibaba’s Hong Kong listed shares rose 5.1% on Tuesday, to their highest in a month, and Tencent gained more than 7%.
Why Alibaba designs its own chips
United States export controls keep Nvidia’s most advanced accelerators out of China. The same controls largely block Chinese firms from using TSMC. Bloomberg reported that it is unclear whether Alibaba works with SMIC, a capacity constrained rival.
Huawei made its own move last week, unveiling new chip technology at its Connect conference.
Neil Shah, vice president at Counterpoint Research, told the AP that using extra computing power to make up for chip limits helps China stay strong locally. His colleague Parv Sharma pointed at manufacturing. How far the gap narrows also depends on China’s ability to advance its own foundries, he said, not just on chip design.
A day before the summit
Xi Jinping arrives in Washington on Wednesday for a state visit and a meeting with Donald Trump. AI, trade and tariffs should all feature. Jensen Huang and Satya Nadella are due at the state dinner.
Treasury Secretary Scott Bessent and Vice Premier He Lifeng met this week. They agreed to set up a dialogue on AI, aimed at a common understanding of goals and threats.
Washington has accused Alibaba and DeepSeek of illicitly accessing American models, a charge Beijing rejects as groundless. Anthropic accused Alibaba of distillation in June.
Dario Amodei, Anthropic’s chief executive, called this month for a slower pace of model development. Wu addressed the argument directly.
Last year, we asserted that the more capable AI becomes, the more powerful humanity will be. Today, I remain steadfast in that conviction.
He calls it machine thinking
Wu avoided the term artificial intelligence through most of his keynote. He used machine intelligence, or machine thinking, instead.
He compared the moment to the industrial revolution. Steam and combustion engines, he said, were built to do what horses and labourers already did, and machine power now drives 99.9% of the world’s physical work.
He expects the same pattern in cognition. Machines currently produce less than 3% of all thinking, by his measure. He predicted they will eventually produce more than 1,000 times the thinking of all humanity combined.
Wu also said the groundbreaking products of the era have not arrived yet. He likened AI coding to the light bulb, an early application of electricity rather than the thing that mattered most.
