Editorial illustration for Deepseek Adapts Open-Source AI Software for Huawei's Ascend Chips
DeepSeek Releases CUDA Alternative for Huawei Chips
Deepseek Adapts Open-Source AI Software for Huawei's Ascend Chips
Deepseek posted the announcement on its official WeChat channel: the Hangzhou AI lab is releasing open-source programming tools built for Huawei's Ascend chips, with Huawei's backing. The centerpiece is TileLang, a language meant to replace Nvidia's CUDA as the entry point for programming AI hardware. Peking University researchers built TileLang originally, and Deepseek has spent roughly a year using it internally, first testing the language on older Nvidia chips before turning it toward Huawei's hardware.
The release goes beyond one language. Deepseek and Huawei also optimized a supernode, a cluster of 128 Ascend 950 chips, and packaged libraries for computation and for moving data between chips, all open source, according to Reuters. The New York Times reports TileLang has become Deepseek's main tool in its work toward artificial general intelligence.
The move matters because chips alone don't win a hardware race. Huawei has poured money into Ascend silicon, but without programming tools that rival CUDA's ease of use, developers have little reason to switch. Deepseek's bet is that a simpler, universal language could get comparable performance out of domestic chips, and that bet now has Huawei's explicit support behind it.
TileLang is at the core of the release. The open-source programming language for AI chips was originally developed by researchers at Peking University, and Deepseek has been using it for about a year. Deepseek argues that anyone trying to build an independent software ecosystem for AI chips first needs a universal language that's easy to program but still gets full performance out of the hardware.
Why this matters
Nvidia's real moat has never been silicon, it's CUDA, the decade of tooling and muscle memory that makes switching chips painful. Deepseek's TileLang push, built on work out of Peking University and battle-tested internally for a year, is a direct attempt to file that moat down for Huawei's Ascend line. If a language layer can make Ascend chips feel as workable as Nvidia's, the calculus for Chinese labs and startups changes fast, especially under export controls that make Nvidia hardware harder to get.
For developers and founders outside China, the thing to watch isn't whether TileLang beats CUDA on benchmarks tomorrow. It's whether an open-source, Deepseek-backed toolchain gets enough contributors and production use to become a genuine second option. Software ecosystems win on adoption, not elegance.
Deepseek shipping this openly, rather than keeping it proprietary, suggests they're betting on exactly that kind of network effect. Worth tracking who else in China starts building on Ascend because of it.
Related Reading
AI Tools & Apps Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps OpenAI Developer: AI Agent "Swarms" Waste Tokens, Create "Coordination Tax
AI Tools & Apps LG's recent webOS update adds Microsoft Copilot app, now removable
From Archives OpenAI expands Codex API with security scans and agents that run software
From Archives Manus 2.0 Adds Event Triggers, Remote Computer Control From Phone
Common Questions Answered
What is TileLang and why is Deepseek releasing it as an open-source tool?
TileLang is an open-source programming language originally developed by Peking University researchers that is designed to replace Nvidia's CUDA as the standard entry point for programming AI hardware. Deepseek has been using TileLang internally for approximately one year and is now releasing it to help build an independent software ecosystem for Huawei's Ascend chips, arguing that a universal, easy-to-use programming language is essential for achieving full hardware performance.
How does TileLang address Nvidia's competitive advantage in the AI chip market?
Nvidia's primary competitive moat has historically been CUDA, the decade-old software tooling that creates significant switching costs for developers and organizations. By offering TileLang as a universal programming language layer for Huawei's Ascend chips, Deepseek aims to reduce this switching friction and make Ascend chips feel as accessible and workable as Nvidia's hardware, potentially changing the calculus for Chinese AI labs and startups.
What role did Peking University play in developing TileLang?
Peking University researchers originally developed TileLang as the foundational programming language technology. Deepseek subsequently adopted and refined the language through approximately one year of internal testing and optimization, first validating it on older Nvidia chips before adapting it for Huawei's Ascend architecture.
Why is Deepseek's TileLang release particularly significant given current export controls?
Under existing export controls that restrict Chinese access to Nvidia chips, TileLang provides an alternative software pathway that makes Huawei's Ascend chips more practical and accessible for Chinese AI development. By lowering the programming barrier to Ascend hardware through a universal language layer, Deepseek enables Chinese labs and startups to develop AI applications without relying on restricted Nvidia technology.
Further Reading
- DeepSeek partners with Huawei to develop chip programming tools, reducing reliance on Nvidia - Reuters
- DeepSeek and Huawei partner on open-source AI chip software - Quartz
- China's DeepSeek open-sources tools to help Huawei chips supplant Nvidia in AI - South China Morning Post
- DeepSeek Teams Up With Huawei To Advance Ascend Chip Infra, Cut Nvidia Reliance - NDTV Profit
- DeepSeek unveils Huawei AI chip tools that may replace Nvidia - Moneycontrol
