DeepSeek AI is Reportedly Running on Huawei’s AI Chips – Is China Moving Away from Nvidia?

The AI competition between the U.S. and China is intensifying, and a new development in DeepSeek AI's infrastructure is adding fuel to the fire. A post by tech tipster Alexander Doria (@Dorialexander) shared an image from Huawei announcing that the distilled version of DeepSeek-R1 is now hosted on the company's AI platform.

Huawei describes its ModelArts Studio as "Ascend-adapted," confirming that the Ascend series chipsets power the data centers. However, the company has not explicitly stated whether these same chips were used to train DeepSeek-R1.

What's Special About Huawei's Ascend 910C?

The Ascend 910C GPU is often described as China's alternative to Nvidia's AI chips, particularly the H800. While it doesn't match Nvidia's top-tier performance, it is powerful enough for AI inference, the process of generating responses from a trained model.

Reports indicate that DeepSeek-R1 was initially trained on Nvidia's H100 GPUs, a common choice for large-scale AI models. However, its deployment on Huawei's Ascend hardware suggests that China is actively shifting toward self-sufficiency in AI infrastructure.

This transition is significant because AI models perform best when trained and deployed on the same hardware. Adapting models to different GPUs can be challenging and time-consuming. If DeepSeek-R1 is running smoothly on Huawei's chips, it could indicate progress in China's AI hardware capabilities, even if Nvidia still leads in raw power.

OpenAI Accuses DeepSeek of Using Its AI Models

As if hardware debates weren't enough, DeepSeek is also facing allegations from OpenAI. The U.S.-based AI firm claims that DeepSeek-R1 may have been trained using OpenAI's proprietary models, though no concrete evidence has been provided.

The lack of transparency in DeepSeek's development fuels these concerns. While DeepSeek has open-sourced its model weights, it has not disclosed its training data or methodologies. Adding to the skepticism, the company claims it trained DeepSeek-R1 for just $6 million, a figure that many AI experts find surprisingly low given the computational demands of such models.

China's AI Strategy and the Future of Huawei's Chips

DeepSeek's move to Huawei's ModelArts Studio is just one piece of a larger puzzle. China's AI sector has been under increasing pressure since the U.S. government banned the export of Nvidia's most powerful AI chips, including the A100 and H100 GPUs. In response, Chinese firms have been ramping up domestic AI hardware development.

Huawei is already working on the Ascend 920C, an upgraded GPU designed to compete with Nvidia's upcoming Blackwell B200 chips. If successful, it could bring China closer to reducing its reliance on Western AI accelerators.

What This Means for the Global AI Race

DeepSeek's growing reliance on Huawei's AI chips highlights China's push for AI independence. While Nvidia still holds the upper hand in performance, China is making strides in developing alternative AI hardware that can support large-scale models.

For now, the biggest question is whether Huawei's AI chips can close the performance gap with Nvidia. If China succeeds in developing competitive AI hardware, it could reshape the global AI power balance in the coming years.

Via