Chinese artificial intelligence start-up DeepSeek open-sourced a suite of six core software modules on Wednesday, all tailored for Huawei Technologies' Ascend AI chips. The release marks a concrete step in China's effort to reduce dependence on processors from US giant Nvidia amid American export controls.

The Hangzhou-based company published the modules through its official WeChat account. The tools mirror DeepSeek's prior open-source releases for Nvidia hardware, with the stated goal of building a new "independent and controllable" software ecosystem for graphics processing units.

The headline release is an Ascend-compatible version of TileLang, a custom programming language designed to streamline development of high-performance kernels — the essential computation programs that run on GPUs and CPUs. TileLang still lists Nvidia as its primary back end, but the project's GitHub page now confirms official support for Huawei's Ascend 950 accelerators, offering "native code generation, automatic scheduling, and synchronisation."

"As an open-source project, the TileLang Ascend version aims to serve as an example for building a highly available software ecosystem for more AI chips," DeepSeek said.

Alongside TileLang, DeepSeek released a set of underlying libraries built for high-throughput training and inference on Ascend silicon. DeepGEMM Ascend is a matrix multiplication kernel library. DeepEP Ascend enables efficient large-scale cross-device communication. Further modules target the efficiency of long-context processing and data filtering on Huawei chips.

DeepSeek said the new tools delivered computational and communication performance approaching "the hardware limits" in several tests. The company did not publish detailed benchmark figures in the announcement.

Joint engineering with Huawei

DeepSeek credited "close collaboration" with Huawei for the Ascend-facing infrastructure. Engineers from both companies worked together on optimisation of a supernode architecture powered by 128 Ascend 950 processors — a configuration that points to large-scale training and inference deployments built entirely on domestic hardware.

The partnership answers Beijing's call for technological self-reliance. When DeepSeek launched its V4 model in April, Huawei quickly announced "full support" with its chips for inference, the process of using a trained AI model to make predictions on new data.

Other Chinese chip designers have followed. Cambricon Technologies, Moore Threads and Alibaba Group Holding's T-Head have all confirmed compatibility with the DeepSeek model. Alibaba owns the South China Morning Post, which first reported the release.

Huawei accelerates its roadmap

Huawei has meanwhile accelerated its own processor cadence. Earlier this month the company said the Ascend 960DT, a chip designed for AI model training, would be "ready in the first quarter of 2027" — three quarters ahead of schedule.

For developers, the open-sourced stack lowers the cost of porting workloads away from CUDA-locked tooling. TileLang's dual-back-end design means a single kernel language can now target both Nvidia and Ascend hardware, reducing the engineering barrier that has historically kept Chinese AI labs tied to US silicon.

The release also signals how DeepSeek positions itself in the ecosystem: not only as a model developer but as an infrastructure contributor. By open-sourcing kernel libraries, communication layers and a programming language back end, the company is effectively publishing the low-level building blocks other Chinese AI firms need to run training and inference on domestic accelerators at scale.