Huawei and DeepSeek release open source AI tools in a bid to lower Nvidia exposure

DeepSeek has released a suite of open-source infrastructure components designed to support Huawei's Ascend AI accelerators. The release includes new repositories like DeepGEMM-Ascend and DeepEP-Ascend, alongside updates to existing tools such as TileLang. These efforts aim to provide a viable alternative to Nvidia's dominant CUDA ecosystem by reducing switching costs for developers. By aligning APIs and buffer interfaces with existing Nvidia-compatible tools, the firms hope to lower the barrier for adopting Chinese hardware. However, adoption remains limited, as evidenced by significantly lower GitHub engagement compared to Nvidia-native counterparts. Furthermore, the tools are currently restricted to specific hardware configurations, such as the Ascend 950 series. While these releases represent a strategic move to bypass US export restrictions and reduce reliance on Western technology, developers continue to face challenges regarding the stability and usability of the Ascend software stack compared to the mature Nvidia ecosystem.