DeepSeek V4 at the center of the US-China chip challenge

global technology industry look at the next move of DeepSeekthe startup cinese has become one of the symbols of Beijing’s ambitions in the race forartificial intelligence.

The sector has been awaiting the possible launch of the new model for weeks V4after the debut, at the beginning of 2025of a low cost conversational agent who he had shown he could compete with Gemini, ChatGPT e Claude.

The crucial issue of chips

At the center of the rumors is above all the question of chips used to train and operate the new system: high-end US processors or Chinese alternatives developed to reduce the technological dependence from abroad.

Second Wei Sunanalyst of Counterpoint Researchcited by the agency Afpunderstanding what hardware infrastructure supports V4 is a key step in evaluating the path forward China towards self-sufficiency in AI.

The Huawei hypothesis

The site The Information reported that V4 could work with latest chips produced by Huawei. It would be a significant step, given that China continues to lack access to H200 on Nvidiadespite the partial easing of export restrictions decided by the United States.

In view of the possible launch, giants such as Alibaba, ByteDance e Tencent they would have already done large orders of Huawei chips.

The birth and rise of DeepSeek

Founded in 2023 as a side project of a hedge fund who had amassed a vast supply of Nvidia processorsDeepSeek shook the markets in January 2025 with the launch of R1.

The chatbot, equipped with advanced reasoning abilityhad triggered a decline in US technology stocks. Donald Trump he had called it a “alarm bell” for the Silicon Valley.

From V3 to V4: a leap in quality

R1 was based on V3the company’s previous major model, introduced in December 2024. According to several observers, V4announced as multimodal modelit might have a even stronger impact.

The system would be able to generate texts, images and videos. Stephen Wufounder of the fund Carthage Capitalbelieves that it will not be a simple software update, but a open source model highly performing ea lower costs compared to competitors.

The challenge of industrial autonomy

For DeepSeek, however, what is at stake is also about industrial credibility. So far the results have been supported by power of Nvidia chipswhile the passage to Chinese semiconductors would require a deep architectural reengineering.

The United States justify blocking the export of the most advanced AI processors with national security reasons. In this context, Wu observes, the delay in the launch of V4 highlights the difficulties in developing advanced models without access to US hardware.

Hypotheses and controversies

Some reconstructions claim that DeepSeek would have succeeded circumvent the prohibitionstraining V4 with thousands of Nvidia Blackwell chips smuggled into China through Third countries.

Meanwhile another Chinese startup, Zhipupresented a image generator which, according to him, he was trained entirely on chip Huawei.

A decisive test for China

According to rumors circulating online, a first attempt by DeepSeek to use i Ascend chip in Huawei would have failed, forcing the company to discreetly return to the Nvidia chips subject to restrictions.

If DeepSeek really succeeded in training V4 entirely on Huawei chipsthe analysts conclude, the result would mark a significant step not only on the plan technologicalbut also in the geopolitical balances of the artificial intelligence sector.