Chinese AI developer DeepSeek has joined forces with Huawei to develop software optimised for the domestic chipmaker's AI accelerators, according to a report from Mobile World Live.

The collaboration centres on adapting DeepSeek's large language models (LLMs) to run efficiently on Huawei's Ascend series of AI processors. Both are subject to US export-control restrictions that limit access to Nvidia's high-end GPUs.

Why are DeepSeek and Huawei working together?

The partnership reflects a broader push inside China to build a self-sufficient AI compute stack. Huawei's Ascend chips have become the leading domestic alternative to Nvidia's H100 and H200 for Chinese firms cut off from advanced US silicon.

DeepSeek has built its reputation on cost-efficient model training. The Hangzhou-based company released its R1 reasoning model in January 2025, drawing industry attention for reportedly matching US frontier-model performance at a fraction of the stated training cost. Running such models on domestic hardware removes a critical dependency on restricted foreign compute.

What does Huawei bring to the table?

Huawei has invested in its own AI accelerator roadmap since US sanctions tightened in 2019. The Ascend line uses Huawei's Da Vinci architecture. The 910C generation is widely cited as competitive with Nvidia's H100 in certain inference workloads.

The chipmaker also developed the CANN (Compute Architecture for Neural Networks) software stack as the alternative to CUDA — Nvidia's proprietary programming environment that dominates AI development. Improving software support for Ascend chips remains the central hurdle for broader industry adoption.

What does DeepSeek contribute?

DeepSeek's expertise in model architecture and training efficiency could help close the performance gap between Ascend hardware and Nvidia systems. The startup has published research on mixture-of-experts models and multi-head latent attention, both of which require careful hardware tuning to deliver competitive tokens-per-second throughput.

What is the market impact?

The collaboration signals continued alignment between frontier-model developers and domestic chip vendors in China. Competitors including Alibaba, Baidu and Tencent have all signalled work to optimise their models for Huawei silicon in recent months.

For Nvidia, the partnership underscores the durability of China's pivot away from US accelerators. Chinese firms collectively placed significant orders for US AI chips before restrictions tightened; subsequent procurement has shifted toward Huawei and other domestic alternatives.

What remains unclear?

Mobile World Live's report did not specify:

• Which Ascend chip generation the partnership targets
• Timeline for production deployment
• Whether the work is exclusive or applies to other chipmakers

The development sits inside a wider Chinese effort to build alternative compute infrastructure. Progress will likely depend on whether Huawei can scale production of advanced Ascend chips amid its own supply-chain constraints.

Neither DeepSeek nor Huawei had issued a public English-language statement on the partnership at the time of the Mobile World Live report.