DeepSeek entrusted the final tuning of its latest model to Huawei, not Nvidia

A shift has occurred in the artificial intelligence industry whose significance extends far beyond a single corporate decision. DeepSeek — a Chinese company whose language models have repeatedly made Silicon Valley nervous over the past year — has entrusted the final optimization of its newest development not to Nvidia, but to Huawei and other Chinese computing accelerator manufacturers. This is reported by Reuters citing informed sources.

To understand the scale of this decision, one needs to understand how the process of creating large language models works. Development is divided into several stages: first, the model is trained on vast arrays of data, then it undergoes additional fine-tuning, and finally — a concluding stage of so-called polishing. It is at this final step that computing accelerator suppliers optimize the model for their hardware architecture, achieving maximum performance and efficiency. Until now, in global practice, this stage has belonged almost exclusively to Nvidia — the company whose GPUs have become the golden standard for training and deploying neural networks. DeepSeek's decision breaks this unspoken monopoly.

Context is critically important here. Since 2022, Washington has consistently tightened export restrictions on advanced chip shipments to China. Nvidia's flagship accelerators fell under the ban — first the A100 and H100, then their "lighter" versions, developed specifically for the Chinese market. Each new round of sanctions narrowed Chinese companies' ability to use American hardware. But in parallel, it created a powerful incentive to develop their own ecosystem. Huawei, despite its own sanctions problems, is actively developing the Ascend accelerator line, which is positioned as a direct alternative to Nvidia's products for machine learning tasks.

The fact that DeepSeek chose Huawei specifically at the final optimization stage speaks volumes. This is not merely a symbolic gesture or a forced measure due to lack of access to Nvidia chips. Final optimization requires deep integration between software and hardware, close interaction between model developers and engineers creating the accelerators. In essence, DeepSeek declares: the Chinese hardware platform has reached a level of maturity where it can be relied upon at the most critical moment — immediately before product release.

For Nvidia, this is a troubling signal, though not catastrophic. Jensen Huang's company still dominates the global market for AI accelerators, and its data center sales revenue continues to set records. However, China is not simply one market. It is the world's second-largest economy and a country with an ambitious state program for artificial intelligence development. Each time a major Chinese player demonstrates viability without American technology, Nvidia's position in this market weakens irreversibly. Even if sanctions are ever relaxed, the habits and ecosystem built around Ascend will not go anywhere.

For Huawei, this is a moment of triumph. The company that Washington tried to cut off from global semiconductor supply chains now serves as a key partner for one of the most promising AI startups on the planet. This strengthens Huawei's position not only as a manufacturer of smartphones and telecommunications equipment, but also as a full-fledged player in artificial intelligence infrastructure. The Ascend line receives a powerful reference case that will work to attract other customers both within China and beyond.

It is worth noting the broader picture as well. DeepSeek has already demonstrated that it can create competitive models with significantly lower computational costs than American competitors. If it now turns out that final optimization on Chinese equipment is not inferior in quality to optimization on Nvidia, this will overturn the notion of how critical dependence on American chips is for the development of cutting-edge AI.

We are witnessing the formation of a parallel technological universe. No less powerful, no less ambitious — simply built on a different foundation. And DeepSeek's decision to entrust Huawei with the final optimization of its model is not simply corporate news. It is a marker that technological divergence between the USA and China has passed the point of no return in yet another critically important domain.