HANGZHOU, China — September 22, 2026 — Alibaba Group is planning a new artificial-intelligence model with between 5 trillion and 10 trillion parameters, according to Reuters reporting from the company’s Apsara Conference in Hangzhou.
The announcement puts model scale, computing infrastructure and AI chips back at the center of the global technology race. Alibaba CEO Eddie Wu said the company’s Qwen team is working toward models designed for more complex, longer-horizon tasks and toward what Alibaba describes as artificial superintelligence.
What was reported
Reuters reported that Alibaba plans to scale its upcoming Qwen 4.5 and Qwen 5 series to as many as 5 trillion to 10 trillion parameters. Alibaba said Qwen 4 is already in training. Parameters are the learned variables inside an AI model and are commonly used as a rough indicator of model size, although parameter count alone does not determine how useful, fast or capable a system will be.
The planned scale is substantially larger than Alibaba’s current flagship Qwen3.8-Max, which Alibaba announced in August with 2.4 trillion total parameters. Alibaba says that model uses a sparse mixture-of-experts architecture and activates a smaller subset of parameters for a given task, illustrating why total parameter count should not be treated as a direct measure of computing cost or real-world performance.
Alibaba also unveiled the Zhenwu V900 AI chip at the conference. Reuters reported that the chip is being developed by Alibaba’s T-Head semiconductor unit, is designed to provide substantially higher performance than the company’s M890 AI supernode, and is planned for mass production in early 2027. Alibaba is also targeting a major expansion of its global data-center capacity, with Reuters reporting a goal of more than 20 gigawatts by 2032.
Why this matters
The important development is not simply the headline parameter number. Training and operating very large AI systems requires a combination of computing hardware, networking, memory, data, software and energy. Alibaba’s announcement therefore points to a broader strategy: building models, chips and cloud infrastructure together rather than treating the model as a standalone product.
Alibaba has been moving in this direction for months. In August, the company reported that AI-related product revenue had recorded triple-digit year-over-year growth for a twelfth consecutive quarter and said its AI cloud and compute services were benefiting from rising demand. Alibaba has also previously described Qwen as part of a full-stack strategy spanning models, cloud infrastructure and proprietary chips.
For readers following the wider AI race, this is another sign that competition is increasingly shifting from individual chatbot launches toward AI infrastructure and agentic systems capable of handling longer, more complicated workflows.
What is still uncertain
The 5 trillion to 10 trillion figure is a stated development target, not evidence that a model of that size has already been completed or that it will automatically outperform smaller systems. Alibaba has not established through this announcement what the final models’ benchmark performance, release schedule, operating cost or availability will be.
It is also important to separate company plans from independently verified performance. Claims about future models should be evaluated again when the models are released and tested under comparable conditions.
Analysis: what to watch next
This section is Valuted News analysis, not a report of a confirmed future outcome.
Three indicators will be especially useful. First, watch whether Alibaba can turn its planned model scale into measurable gains on reasoning, coding, multimodal understanding and long-horizon tasks. Second, watch the availability and economics of the Zhenwu hardware, because large models only become commercially meaningful when the underlying compute can be supplied at practical cost. Third, watch how Qwen’s open and commercial ecosystem develops, including adoption by developers and businesses.
Readers who want broader context can also see our AI safety coverage and our global markets and AI outlook. For a broader guide to following major international developments, see Valuted News World News.
Bottom line: Alibaba has announced an ambitious plan to push future Qwen models into the 5 trillion–10 trillion parameter range while expanding its AI-chip and data-center strategy. The scale is significant, but the real test will come later: whether the resulting systems deliver measurable capability improvements at a cost and infrastructure footprint that businesses and developers can actually support.
Sources: Reuters, September 22, 2026, reporting from Alibaba’s Apsara Conference; Alibaba Group, August 3, 2026, Qwen3.8-Max announcement; Alibaba Group, August 20, 2026, quarterly AI and cloud update. Valuted News distinguishes reported facts from its own analysis and does not treat company projections as established outcomes.