🇨🇳 DeepSeek opens Ascend stack

Chinese AI firm DeepSeek said Wednesday it open-sourced infrastructure components for Huawei’s Ascend computing platform. The release covers the TileLang high-level compiler, a compute library, and a distributed communication library, mirroring its previously open-sourced components for Nvidia platforms.

DeepSeek said TileLang is simpler than Nvidia’s CUDA for programming and can better utilize chip features; it already supports most operator implementations used in training the DeepSeek V4 model series. The move comes as DeepSeek increases adoption of Huawei and other domestic chips, as part of a broader push to reduce reliance on Nvidia amid US advanced chip export controls.

DeepSeek still depends on Nvidia chips for model training, but has reportedly prioritized Huawei chips for future training workloads. US media previously reported the company expects a new batch of Huawei training chips in late 2026 or early 2027, though model and volume remain unclear.

#China

@asianomics