Published 14 hours ago • loading... • Updated 2 hours ago
DeepSeek partners with Huawei to develop chip programming tools, reducing reliance on Nvidia
The Chinese startup said Huawei gave full support as the two companies built software for 128 Ascend 950 chips and a supernode system.
On Wednesday, Chinese AI start-up DeepSeek open-sourced core tools tailored for Huawei Technologies' Ascend AI chips, aiming to reduce reliance on U.S. giant Nvidia processors amid American export curbs.
DeepSeek released six software modules to build an "independent and controllable" software ecosystem for GPUs, reflecting accelerated efforts within China's tech sector to establish self-controlled supply chains amid tightening U.S. export controls.
Among these releases is an Ascend-compatible version of TileLang, a custom programming language for kernel development. The project now officially supports Huawei's Ascend 950, offering "native code generation, automatic scheduling, and synchronisation," according to GitHub.
DeepSeek stated "TileLang was created precisely to meet this need," offering "a simpler programming model" than Nvidia's CUDA. The language improves development efficiency and simplifies code logic for developers.
By bridging the gap between advanced large language model architectures and domestic hardware, DeepSeek and Huawei are laying groundwork for a robust alternative to Western computing dominance. The partnership reflects China's broader push for semiconductor independence.
Chinese DeepSeek partnered with Huawei Technologies to develop optimized programming tools for Huawei Ascend chips, deepening collaboration between two of China's leading technology companies. Exclusive material for subscribers. To have full access, access the material link and register.