The United States and China are immersed in a trade and technological war that has caught the line of fire to the AI giant: Nvidia. The situation is that Nvidia must prioritize AI companies from the United States to guarantee the supremacy of this country, but as a company it would be interested in taking a bite out of the giant Chinese market. And the problem is twofold: it has not been able to do so for a long time due to trade vetoes, but now that it seems that it can sell its famous H200 to China, it turns out that China has turned the page.
More or less.
green light. Nvidia has gone from having a monopoly on AI GPUs in China to have a 0% quota. These are the words of the CEO, Jensen Huang, and the reason is the aforementioned trade restrictions between the powers that prevented Nvidia from selling its most powerful products to the Asian giant. Huang has spent months insisting on Donald Trump’s government to allow them to sell with a very clear logic: China is going to develop its alternatives and what better way to make a profit until then.
The situation is gone relaxing at the end of last year and at the beginning of this to get to the point where we are now. According to Reutersthe US Department of Commerce already allows ten Chinese companies and distributors such as Foxconn and Lenovo acquire that long-awaited H200the company’s second most powerful AI chip. Good news for the company. Or they should be if it weren’t for the fact that the Chinese industry is going its own way
looking home. Alibaba, ByteDance, JD.com and Tencent are the Chinese giants that can supposedly already buy H200. Up to 75,000 chips each, to be exact. However, it is noted that they have not yet made any shipments. Here there is a mix between very restrictive bureaucracy and, above all, that emphasis on national development. Tencent, for example, noted in September last year that they had no intention of producing AI chips, but that they were going to invest a lot of money in domestic partners.
For example, they are in the process of adapting their infrastructure to be able to connect Huawei’s Ascend platform (particularly the Ascend 950 series) as the main training tool for large models. A few days ago, Tencent’s strategy director already pointed out that that strategy was still in place and that the company expects a significant increase in spending on AI GPUs designed in China.
Manufacturing at home. Alibaba and Bytedance have a different approach. If Tencent is focusing on acquiring Huawei platforms, Alibaba and Bytedance are looking to create their own chips. Alibaba seeks to be the most powerful RISC-V chip created to date and it was reported that Bytedance wanted Samsung will manufacture its processor.
In the end, whether buying from Huawei or developing the tool internally, the two approaches respond to the great national objective: that at least 50% of the data centers that belong to the State use at least 50% Chinese integrated circuits in their servers. That is one of the great Chinese technological impulses of recent years, one of the crucial points of the Five-Year Plan for the development of the country and, above all, the strategy that Nvidia had been warning the United States about for some time.
The age of inference. Because this period of ostracism to which the US condemned China has served for the country to develop three very clear alternatives to Nvidia and encourage companies that are already working with models to develop their own hardware. This is important especially in the new AI framework we are entering, that of inference.
Although the AI will continue to train and GPUs will be needed for this, the next step is inference, the agentic era in which the processor or CPU is very important. AMD is moving there, same as Intel or ARMand precisely processors are something that Huawei is good at and in which the Chinese giants can shine as much as their American counterpart by developing chips tailored to their models and needs.
Also, as pointed out in CNBChaving your own chips means you don’t have to fight with anyone else in a time when there is scarcity and, of course, if you don’t have to buy from an outsider, there is an improvement in the gross income margin.
juicy cake. And this leaves Nvidia in that uncomfortable situation, one in which it wants to participate, but in which it seems that it is no longer needed as much as before. Because China is developing its chips for this new era of AI and Nvidia is running into a final boss called bureaucracy and the pressure groups of the ‘Make America Great Again‘.
The first is due to the slowness of the export order processes, something that takes months when orders should be much more agile. The second are the aforementioned pressure groups that hold that any deals Nvidia makes with Chinese companies are less chips for American companies, something that should not be allowed.
Meanwhile, Chinese companies are developing their alternatives and Huawei wants to flood the market with 750,000 chips this year, three times more than its shipments in 2025, and Nvidia is falling short of a $50 billion pie.
In Xataka | The US has the best AI models. China has something else: AI too cheap to care about
GIPHY App Key not set. Please check settings