DeepSeek launched open-source software on Wednesday, September 30, to help developers program Huawei’s artificial intelligence chips. This toolkit offers a homegrown alternative to Nvidia’s dominant CUDA platform. DeepSeek said, "A powerful programming language that stays simple to code in was essential for building a healthy AI ecosystem" for Chinese developers.
China's DeepSeek launches open-sourced software with Huawei; May rival Nvidia's CUDA
The Chinese startup planned to install at least 160,000 of Huawei's most powerful accelerators at a large data centre it was building in Inner Mongolia a Bloomberg report said in August.
By CNBCTV18.com
DeepSeek launched an open-source free software on Wednesday, September 30, that lets developers program Huawei's artificial intelligence chips, giving China a homegrown alternative to the Nvidia tools that dominate the global industry.
The Chinese startup made its toolkit for Huawei's Ascend accelerators open-source, meaning anyone can download, use and modify it, DeepSeek said on its WeChat account. The move offered Chinese developers a way to build AI without depending on American technology.
The package includes TileLang, billed as China's answer to Nvidia's CUDA, the software platform regarded as the worldwide standard for building AI. Developers use such platforms to write the instructions that tell AI chips what to do.
DeepSeek built open-weight models, which release their core settings publicly, to rival those from Anthropic and OpenAI, while Huawei led the country's push to make the accelerators that power data centres.
The startup said, in its WeChat post, that a powerful programming language that stays simple to code in was essential for building a healthy AI ecosystem. The company said it had worked with Huawei to boost performance on the Ascend 950 chips.
Huawei gave "unreserved and vigorous support" throughout research and development, and that the two would work together on further innovations, DeepSeek added. The startup already used domestic chips, although most AI model training worldwide still relied heavily on Nvidia's semiconductors.
The startup planned to install at least 160,000 of Huawei's most powerful accelerators at a large data centre it was building in Inner Mongolia, according to a Bloomberg report.
The company was also finalising a fundraising round that valued it at close to 500 billion yuan ($74 billion), ahead of a public listing that could come as early as this year, a separate Bloomberg report said.