The US has given the green light for the sale of Nvidia chips to 10 Chinese companies. But Beijing is holding back for both geopolitical and strategic reasons: it already has its own path to AI development that does not necessarily rely on computing power by Andrea Monti and Arcangelo Rociola – Originally published in Italian Tech – La Repubblica
On the sidelines of his summit with Xi Jinping, Donald Trump announced that he had authorised the sale of Nvidia H200 chips to ten Chinese companies, including Alibaba, Tencent and ByteDance, for a maximum of 75,000 units each. There is a catch. Beijing – according to an exclusive report by Reuters – has not yet given the green light for the purchases. And it is not certain that it will.
The block comes from the Chinese government, which is holding back so as not to weaken the development of the domestic semiconductor industry, led by Huawei. Complicating matters, Washington is demanding guarantees that the chips will not be used for military purposes and has imposed stringent commercial conditions: 25% of sales revenue would go back to the US and the chips would have to physically pass through US territory. Beijing fears that the devices may harbour vulnerabilities or backdoors. Anti-China hawks in the Trump administration remain opposed to any agreement.
The surprise arrival of Nvidia’s CEO. But the two powers are at a standstill
Also present at the summit was Jensen Huang, CEO of Nvidia, who was included at the last minute in the American delegation at Trump’s direct invitation. Before the export restrictions, Nvidia controlled 95% of the Chinese market for advanced chips. China accounted for 13% of its total revenue. Huang estimates that the Chinese AI market will be worth $50 billion by 2026 alone. These figures clearly illustrate the stakes involved. With specific developments and changes in recent months that have overturned the initial outlook.
Artificial intelligence processors are one of the most significant issues currently under discussion between the US and China. The two nations are at an impasse: on the one hand, the United States needs rare earths; on the other, China needs access to the latest GPUs and lithography technologies for processor manufacturing. This is the crux of the technological dispute between the two superpowers. Mediation on this issue, assuming it is actually possible, must be viewed within the broader strategic frameworks of the two countries.
US: technology embargoes. China: manufacturing autonomy
On the one hand, the United States, through technology embargoes, has sought to limit China’s ability to develop autonomous capabilities comparable to those of the US in the field of artificial intelligence.
On the other hand, China has accelerated the development of its own autonomous technologies across the entire production chain, going so far as to produce its own processors, which, although not entirely comparable to US ones, are nonetheless highly performant.
It has also launched research into alternative approaches, such as those involving analogue processors, which allow for highly efficient results without necessarily using the GPU technologies currently developed by Nvidia. And this offers another possible explanation for the coolness with which Beijing has greeted the new American overture.
DeepSeek is a groundbreaking AI development model that changes everything
When it comes to artificial intelligence models, too, China has adopted an efficiency-oriented strategy. The first versions of DeepSeek demonstrated that it was possible to produce an extremely capable model without the size and complexity of its American counterparts.
If these considerations are broadly correct, the lesson to be drawn is that the approaches historically employed to use technology as a lever for maintaining a strategic advantage over other countries no longer work against nations that have developed a fully autonomous capacity for research, technology transfer and industrial production.
It therefore becomes inevitable to imagine that the US and China will have to reach a sort of armed truce, at least for the time being. Whatever the outcome of the negotiations, it is once again clear that artificial intelligence is a two-horse race between the US and China, in which the European Union cannot even play the role of referee.
Brute computing power vs intelligent training
To this must be added that the new versions of DeepSeek are increasingly less resource-intensive in terms of hardware, and in particular GPUs. This means two things: on the one hand, the development, at the software level, of a totally different approach that does not rely exclusively on brute force and processor capacity; on the other, the real possibility of implementing artificial intelligence systems even in devices which, due to their size constraints, would not have been able to accommodate a sufficient number of GPUs. This effectively removes any technological restrictions, paving the way for a more fluid and effective approach.