DeepSeek is preparing a new AI model that will mark a significant milestone for Huawei, according to The Information. The development signals a deepening alliance between China’s most talked-about AI lab and its most important chipmaker.
The details are thin, but the implications are massive. DeepSeek has already proven it can build competitive AI models using Huawei’s Ascend chips instead of Nvidia’s GPUs. A new model optimized for or trained on Huawei hardware would validate China’s push toward semiconductor self-sufficiency in a way that no benchmark score ever could.
Why this matters right now:
- U.S. export controls are tightening. Washington has spent two years trying to choke off China’s access to advanced AI chips. If DeepSeek keeps shipping competitive models on domestic hardware, those restrictions look increasingly like a speed bump, not a wall.
- Huawei needs a flagship AI customer. The Ascend 910B/C chips are technically capable, but adoption has been the bottleneck. DeepSeek publicly building on Huawei silicon is the best advertisement money can’t buy.
- DeepSeek already shocked the industry once. Their R1 model in January 2025 demonstrated that you don’t need the latest Nvidia H100s to train frontier-class models. A follow-up model doubles down on that thesis.
The bigger picture. China’s AI ecosystem has been quietly reorganizing around domestic supply chains. Huawei’s chip division, Alibaba’s cloud infrastructure, and labs like DeepSeek are forming a parallel stack that doesn’t depend on American technology. Each new model trained on Ascend chips makes that stack more credible.
For Western AI companies and policymakers, this is the scenario they’ve been worried about: export controls accelerating China’s self-reliance rather than slowing its AI progress. DeepSeek’s willingness to optimize for Huawei hardware, rather than hoarding whatever Nvidia chips it can get, suggests the decoupling is already happening by choice, not just by force.
What to watch next: the model’s actual performance benchmarks, which Huawei chips were used, and whether other Chinese AI labs follow DeepSeek’s lead in going fully domestic on compute.
More details are available in the original report from The Information.