June 19, 2026, (Inside AI) — China's leading AI developers are aggressively advancing trillion-parameter foundation models, fueled by falling costs and strong investor backing, even as US export controls tighten.
The trillion-parameter race heats up in Beijing
Chinese firms are rapidly shifting from billion-parameter models to architectures with over a trillion parameters. A Donghai Securities report notes the move includes million-token contexts and full adaptation to domestic chip stacks.
DeepSeek launched its first trillion-parameter model, V4, in late April. Alibaba crossed the threshold earlier with Qwen-3-Max-Preview in September. Xiaomi also joined the trillion-parameter club in recent months.
Washington's clampdown reshapes the landscape
The push comes as the Trump administration imposed new export controls. Last week, Anthropic suspended global access to Mythos and Fable, both estimated to have trillions of parameters, following the restrictions.
The move underscores a deepening US-China tech divide. Parameters are a key measure of AI capability, and Chinese firms aim to close the gap with US leaders like OpenAI and Anthropic.
Domestic chips and lower costs fuel ambition
Chinese developers are adapting models to domestic hardware, reducing reliance on restricted US technology. Falling computational costs and strong investor momentum are accelerating the trend.
Industry analysts say the shift to trillion-parameter models marks a strategic pivot from the general billion-parameter models popular in 2023 and 2024.
What the experts are watching
While Chinese firms tout progress, independent verification of model performance remains limited. Some researchers question whether parameter count alone drives real-world utility.
Yet the scale of investment signals a long-term commitment. Alibaba, which owns the South China Morning Post, and others are betting that massive models will unlock new enterprise applications.
The road ahead
The US controls may slow but not stop China's AI ascent. Domestic chip advancements and open-source collaborations could further narrow the gap. The global AI race is entering a new, more fragmented phase.
