Shedding Dependency: DeepSeek Rolls Out V4 Model for Huawei Chips

The US continues to choke China with sanctions on NVIDIA chips, but Beijing is adapting. On April 24, 2026, the Chinese AI lab released `DeepSeek-V4 Preview`, available in V4-Pro and V4-Flash versions.

The model boasts a colossal context window of 1 million tokens, allowing entire books or source code bases to be loaded into the prompt. But the main macro signal lies in the architecture: the algorithms are natively adapted for Chinese Huawei Ascend neural processors. This is a fundamental shift. Until now, the development of top LLMs has been tightly bound to the NVIDIA CUDA ecosystem. Optimizing flagship models for domestic hardware proves that the technological division of the world into two isolated AI camps is in full swing.

Source: DeepSeek / Reuters