DeepSeek’s collaboration with Huawei Technologies is strengthening China’s AI ecosystem and closing the performance gap with US technology, San Francisco-based research firm SemiAnalysis found in a new study.

The firm cited DeepSeek’s recent release of software tools tailored to Huawei’s Ascend AI hardware, including six software modules similar to ones DeepSeek had previously created for Nvidia chips.

The releases included an Ascend-compatible version of TileLang, a custom programming language designed to simplify the development of high-performance software kernels.

DeepSeek said at the time the Ascend version of TileLang would help promote a more independent and controllable domestic software ecosystem for GPUs.

The Hangzhou-based start-up has also provided Huawei with early access to its models, enabling Huawei to offer immediate inference support for models such as V4 and V4.1 upon release.

Such collaborations mean DeepSeek is likely to rely primarily on Huawei’s already “mature and well developed” technology stack by next year or 2028, SemiAnalysis said.

It added that the companies’ work is driven partly by “external and internal pressure to build a fully domestic Chinese ecosystem”.

The company also cited Huawei’s own engineering progress.

While the Ascend 950PR, announced in March, is still behind Nvidia’s H100, released in 2022, SemiAnalysis noted that the new Huawei chip kernels are achieving efficiency at 99.7 percent of their theoretical ceiling.

The study referenced a remark in April by Nvidia chief executive Jensen Huang that it would be a “horrible outcome” if DeepSeek optimised its technology for Huawei architecture.

“From the software results so far and testing on 950PR, the ecosystem gap is definitely closer with DeepSeek’s help on top of Huawei’s work. Jensen’s nightmare is starting to come true,” the study said.