Alibaba, ByteDance and DeepSeek Race to Build AI Bigger Than Any Before

On September 22, Alibaba's CEO Eddie Wu told an audience at the company's Yunqi conference that Alibaba is preparing to train a far larger AI model. Reports circulating online suggest Alibaba, TikTok-owner ByteDance, and DeepSeek are all aiming at the same target: 10 trillion "parameters." Parameters are the internal dials an AI adjusts while learning. More dials generally means a model that can handle harder problems.

To put the size in perspective, most leading chatbots today are believed to run on a few hundred billion to a couple of trillion parameters. Ten trillion would be a step change — closer to the scale some researchers guess the human brain works at, though that comparison is rough. In practice, a bigger model can hold more context, follow longer chains of reasoning, and handle messy real-world tasks like reading contracts or fixing code with fewer mistakes.

For you, the interesting part isn't bragging rights — it's price. Chinese labs have spent the past year releasing surprisingly capable models for cheap or free, forcing OpenAI, Google and Anthropic to cut their own prices. If this next wave lands, expect AI tools that are smarter and cheaper, better at Chinese-language and local tasks, and increasingly running on your phone rather than only in distant data centers.

The catch: there is no release date, no product, and the news comes from conference remarks and social media chatter, not published research. Bigger is not automatically better — training runs at this scale can cost hundreds of millions of dollars and enormous electricity, and past giant models have disappointed. Think of it as three companies announcing they're building taller skyscrapers. Impressive, but we won't know if anyone can actually live in them for a while.