DeepSeek closes in on US rivals despite chip curbs

The report suggests that Chinese AI developers are finding ways to improve model performance even under chip export controls. Rather than relying solely on access to the most advanced US chips, companies like DeepSeek appear to be optimizing existing hardware through software innovation. Memory efficiency, for example, allows larger or more complex models to run on less powerful or less abundant hardware, directly addressing a constraint imposed by chip restrictions.

Deepening work with Huawei indicates a strategy of leveraging domestic hardware and software ecosystems. This collaboration could involve joint development, specialized hardware-software integration, or optimization for Huawei's AI chip architectures. Such partnerships are crucial for building competitive AI capabilities when access to global supply chains for certain components is constrained.

The reported narrowing of the performance gap between Chinese and US AI models suggests these domestic strategies are yielding results. While specific metrics for this performance gap are not detailed, it implies progress in areas like model accuracy or the ability to handle complex tasks. This development challenges the assumption that chip restrictions would indefinitely widen the gap, showing that software and ecosystem development can offset hardware limitations to a degree.