---
format: "aidr-story-markdown/v1"
id: "2f5100c44c76bb8e165803014d6634cd0b7d93254999420e86415c85880969d1"
canonical_url: "https://aidr.today/2f5100c4?lang=en"
title: "DeepSeek Open Sources AI Toolset to Make Huawei Ascend Viable for Training"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-30T03:07:47.000Z"
category: "Infra"
topics: ["deepseek","open-source","llm"]
source_urls: ["https://marketbrief.now/ai/deepseek-open-sources-ai-toolset-to-make-huawei-ascend-viable-for-traini-8d03a4b7","https://huggingnews.com/ai/deepseek-open-sources-ai-toolset-to-make-huawei-ascend-viable-for-traini-8d03a4b7","https://the-decoder.com/chinas-ai-industry-closes-ranks-as-deepseek-ships-open-source-software-for-huaweis-ascend-chips/","https://x.com/zheanxu/status/2105119944556110265","https://x.com/Reuters/status/2105216087084216641","https://x.com/kyleichan/status/2105432589595259156","https://x.com/kyleichan/status/2105432311059620261","https://x.com/poezhao0605/status/2105195692067020995"]
summary: "A set of high performance kernels and communication libraries now allows developers to train models on Huawei Ascend hardware using a stack previously optimized for Nvidia. DeepSeek open-sourced this infrastructure—comprising TileLang, DeepGEMM, DeepEP, TileKernels, FlashMLA, and DeepSelect—to enabledomestic AI development in China. In official tests for Dense GEMM, the DeepGEMM Ascend library reached 99.8% of the theoretical hardware limit and 98% on MegaMoE. The release was developed in collaboration with Huawei for the Ascend 950 128 card supernode, with deep optimizations for computation and communication. TileLang, a centerpiece of the suite, is a high-level language that permits writing operators in a Python-like format that compiles for Nvidia, AMD, and Ascend hardware. DeepSeek utilized TileLang to implement a large number of operators during the training of its V4 model series."
---

# DeepSeek Open Sources AI Toolset to Make Huawei Ascend Viable for Training

> [Open the canonical story](<https://aidr.today/2f5100c4?lang=en>)

**Published:** 2026-09-30T03:07:47.000Z
**Category:** Infra
**Topics:** deepseek, open\-source, llm

## Summary

A set of high performance kernels and communication libraries now allows developers to train models on Huawei Ascend hardware using a stack previously optimized for Nvidia\. DeepSeek open\-sourced this infrastructure—comprising TileLang, DeepGEMM, DeepEP, TileKernels, FlashMLA, and DeepSelect—to enabledomestic AI development in China\. In official tests for Dense GEMM, the DeepGEMM Ascend library reached 99\.8% of the theoretical hardware limit and 98% on MegaMoE\. The release was developed in collaboration with Huawei for the Ascend 950 128 card supernode, with deep optimizations for computation and communication\. TileLang, a centerpiece of the suite, is a high\-level language that permits writing operators in a Python\-like format that compiles for Nvidia, AMD, and Ascend hardware\. DeepSeek utilized TileLang to implement a large number of operators during the training of its V4 model series\.

## Sources

- [Story source](<https://marketbrief.now/ai/deepseek-open-sources-ai-toolset-to-make-huawei-ascend-viable-for-traini-8d03a4b7>)
- [Story source](<https://huggingnews.com/ai/deepseek-open-sources-ai-toolset-to-make-huawei-ascend-viable-for-traini-8d03a4b7>)
- [Story source](<https://the-decoder.com/chinas-ai-industry-closes-ranks-as-deepseek-ships-open-source-software-for-huaweis-ascend-chips/>)
- [Story source](<https://x.com/zheanxu/status/2105119944556110265>)
- [Story source](<https://x.com/Reuters/status/2105216087084216641>)
- [Supporting source](<https://x.com/kyleichan/status/2105432589595259156>)
- [Supporting source](<https://x.com/kyleichan/status/2105432311059620261>)
- [Supporting source](<https://x.com/poezhao0605/status/2105195692067020995>)