Alibaba Plans Qwen Model Reaching 10 Trillion Parameters
Alibaba says future Qwen models could reach 5–10 trillion parameters as it links model scaling to domestic chips, cloud capacity and agent deployment.
Alibaba has outlined plans to train a future Qwen model with between 5 trillion and 10 trillion parameters, a major escalation in the company’s long-term AI ambitions. Chief executive Eddie Wu made the announcement at Alibaba’s Apsara Conference in Hangzhou, where the company also presented new AI chips, cloud infrastructure and agent-focused products.
A full-stack bet
The proposed model would be several times larger than Alibaba’s current Qwen3.8-Max, which is reported to contain about 2.4 trillion parameters. The announcement concerns a future training target, not a released model: Alibaba has not provided a launch date, architecture, training dataset, compute requirement or expected benchmark results.
That distinction matters because parameter count alone does not determine capability. Sparse architectures, routing efficiency, training data quality and inference economics will shape whether a model of this scale produces a meaningful improvement or simply creates a larger engineering burden.
Alibaba presented the model roadmap alongside its Zhenwu V900 AI chip, which the company described as substantially faster than its previous generation. It also discussed expanding data-centre capacity and building an agent-oriented cloud platform. Together, the announcements suggest that Alibaba is treating frontier models, silicon, cloud services and applications as one connected system rather than separate businesses.
Why it matters
The plan is significant because it is one of the clearest public signals that a major Chinese technology company intends to compete at the very largest training scales while reducing dependence on imported infrastructure. It also raises the strategic stakes for open model development: Qwen’s previous releases have circulated widely among developers, but Alibaba has not said whether a 5–10 trillion-parameter successor would be open-weight.
The immediate story is therefore not a new model available to users. It is a commitment of capital, chips and organisational capacity that could reshape the next phase of China’s frontier-model race.