DeepSeek and Huawei Develop Open-Source Tools for Ascend Chips

DeepSeek says it partnered with Huawei to develop programming tools for Huawei’s Ascend chips. [1] The work includes TileLang, which Techmeme’s Reuters summary describes as an open-source alternative to the CUDA programming platform. [1] Newsquawk separately reported on September 30, 2026, that DeepSeek released open-source programming tools powered by Huawei Ascend chips. [2] Newsquawk placed the release within a recurring pattern of Chinese AI development routing around export controls on advanced foreign semiconductors, while saying the key historical question is whether domestic silicon can operate at scale for training and inference workloads. [2]

The immediate significance is software enablement: DeepSeek is pairing open-source programming tools with Huawei’s Ascend hardware rather than presenting a new model. TileLang’s positioning as a CUDA alternative makes CUDA the clearest competing platform referenced by the sources, while Newsquawk’s context places the effort against constraints on access to advanced foreign semiconductors. The practical test is therefore not simply whether developers can obtain the tools, but whether Ascend systems become usable at scale across training and inference. What the sources don't address: how TileLang performs, which workloads it supports, or how broadly developers have adopted it.