In the world of artificial intelligence, GPUs, i.e. graphics processors, have become a key element for training and executing machine learning models. To date, Nvidia has dominated this sector with its series of advanced chips, such as the H100, and a proprietary software ecosystem, CUDA, which has made its hardware almost irreplaceable. However, the growing technology competition between the US and China is accelerating the search for alternatives, and one of the most promising comes from Huawei, backed by DeepSeek, the Chinese artificial intelligence company that recently shocked the AI world with the release of its new DeepSeek V3.

DeepSeek has announced that DeepSeek V3 will be compatible with Huawei’s Ascend chips as of now, a strategic choice aimed at reducing dependence on American technology. At the heart of this challenge is the Huawei 910C processor, designed to compete with the Nvidia H100, the reference chip for artificial intelligence. While the H100 is considered the industry leader in performance and stability, the 910C offers significant power, with both training and AI pattern inference capabilities. This means that it can be used both in the development phase, where the model is ‘trained’ with huge amounts of data, and in the practical use phase, when the model is ready to be applied in real-world scenarios, such as speech recognition, image generation or analysis of complex data.

Nvidia’s strength: hardware and software in a closed ecosystem

Nvidia’s real advantage, however, lies not only in the power of its GPUs, but in its software ecosystem. CUDA, the platform developed by Nvidia, allows developers to optimise their applications extremely efficiently and make the most of the GPU’s hardware capabilities. Until now, this has been a barrier to entry for any competitors, because migrating entire artificial intelligence systems to hardware other than Nvidia’s would have been extremely complex and expensive. Huawei has found a solution to this problem by developing CUNN, an alternative to CUDA that allows code conversion with a simple command. In other words, AI applications that run on Nvidia can be quickly adapted to run on Huawei chips, significantly reducing the software compatibility problem.

Initial experiments have shown that the Huawei 910C achieves around 60 per cent of the performance of the Nvidia H100 in inference, i.e. in practical use of the trained models. However, DeepSeek’s engineers are working on manually optimising the kernels, i.e. the most critical functions for managing computation, with the aim of further improving performance. This could lead to a narrowing of the gap between Chinese and American solutions.

The technology race between China and the US

The impact of this development goes beyond mere technological confrontation. The race for artificial intelligence chips has become a major strategic and geopolitical issue. The US has imposed severe restrictions on the export of advanced GPUs to China, fearing that these technologies could be used not only to develop new civil applications, but also to enhance the Chinese government’s military and surveillance capabilities. As a result, companies such as Huawei and DeepSeek have no choice but to develop their own infrastructure to reduce dependence on US suppliers and ensure the continuity of their projects.

If this strategy succeeds, we could see a significant shift in the AI chip market. Today, Nvidia is the undisputed industry leader, but if the Chinese industry succeeds in producing competitive hardware, companies around the world may start to consider cheaper alternatives that are less subject to US trade restrictions. For now, Nvidia maintains a distinct advantage, especially in model training, where Chinese chips have not yet achieved the same stability and reliability. But if Huawei and DeepSeek manage to close this gap, we could see the beginning of a new chapter in the global race for artificial intelligence, with an increasingly fragmented market dominated no longer only by American companies, but also by Chinese players ready to challenge the Western technological monopoly.