Alibaba is planning a next-generation AI model with between 5 trillion and 10 trillion parameters, according to CEO Eddie Wu.

The announcement was made at Alibaba's 2026 Apsara Conference as part of a broader push involving AI models, chips and data centers.

If built at that scale, the model would represent a major increase over Alibaba's current flagship systems.

What are parameters?

Parameters are numerical values learned during model training.

They are often used as a rough indicator of model size.

More parameters can provide additional capacity, but bigger does not automatically mean better.

Model architecture, training data, inference methods, reinforcement learning and system design all matter.

Why is Alibaba going bigger?

Alibaba says its next-generation model is intended to support increasingly complex and long-horizon tasks.

This fits into a broader shift in AI development.

Earlier models focused heavily on generating text and answering questions.

Newer systems are increasingly designed to reason, use tools and complete multi-step workflows.

These tasks can benefit from greater model capacity.

But there is a catch

Larger models are expensive.

Training a model with trillions of parameters requires enormous compute, memory, networking and energy.

Inference can also become expensive if every request requires the full model.

That is why the AI industry is increasingly exploring techniques such as:

Mixture-of-experts architectures

Model distillation

Quantization

Dynamic routing

Smaller specialist models

Reinforcement learning

Alibaba's announcement therefore represents both a model race and an infrastructure race.

The company is simultaneously developing the Zhenwu V900 chip and targeting more than 20 GW of data center capacity by 2032.

The bigger takeaway is simple.

The next generation of AI may be defined not just by model size, but by how effectively companies can turn enormous models and infrastructure into useful, affordable intelligence.