DeepSeek unveils new AI model, breaking reliance on US chips

Chinese startup DeepSeek, which surprised the world last year with an affordable yet powerful AI model, launched its new V4 version. Notably this time: the model is designed to run on chips manufactured by Huawei, a Chinese company, instead of US Nvidia chips.

Why is this important?

The United States has restricted the export of advanced Nvidia chips to China since 2022. This forced Chinese companies to seek local alternatives. Huawei responded with its Ascend chips, and DeepSeek was the first company to demonstrate that these chips can train competitive AI models at a fraction of the cost.

DeepSeek's V4 model is the next step in that direction. It not only trains on Huawei chips — the model is also optimized for China's software ecosystem, further reducing dependence on US technology.

The geopolitical context

This move comes amid the trade war between the United States and China. Washington has tried to limit China's access to advanced chips to slow its AI development. China's response — exemplified by DeepSeek — is that those restrictions may accelerate the creation of local alternatives instead of slowing them down.

For companies using or buying AI services, this means the provider landscape is diversifying: it's no longer just the United States and its allies controlling the most powerful infrastructure.