Details
The conflict between the US and China has been growing for some time. The underlying reason is that Washington’s foreign and economic policy since the Obama administration has been geared toward constraining China’s growing power.
Jan Turowski is the head of the Rosa-Luxemburg-Stiftung office in Beijing.
President Donald Trump provoked a trade war with China during his first term, and Joe Biden largely continued it. At the start of his second term, Trump then escalated the conflict by announcing far-reaching tariffs. It is by no means coincidental that these tariffs targeted China in particular; Trump has occasionally threatened Beijing with tariffs of 145 percent. But unlike the tariff conflict during Trump’s first term, this time Chinese leaders have not backed down. Rather, they have responded to each new tariff with tariffs of their own. This approach has been highly successful due to the fact that Washington not only overestimated China’s dependence on the US market but also underestimated the dependence of American service providers, energy companies, and especially agricultural producers on the Chinese market.
Furthermore, China has its own leverage in the form of its global dominance in the mining and processing of rare earth materials. The export restrictions imposed by President Xi Jinping eventually brought the US government to the negotiating table. At their October 2025 summit in Busan, South Korea, Trump and Xi agreed to a one-year “truce”: the tariffs were largely revoked and rare earth shipments resumed.
The Fight Over Chips
As of now, the US government’s trade war appears to have failed. Consequently, the focus is shifting to the technology war being fought over semiconductors. Here, Washington’s strategy is oriented toward securing US firms’ long-term monopoly over artificial intelligence — and thereby a geopolitical tool for the global dominance of the US government.
Both the first Trump administration and the Biden administration imposed significant restrictions on exports of the most advanced chips to China. There is an assumption in Washington that China cannot catch up with the US if it cannot import the most up-to-date semiconductors. Gina Raimondo, the US Trade Minister during the Biden administration, said that the US government would do everything it could “to deny China the most advanced technology”.
However, Washington appears to have miscalculated, as illustrated by recent developments at the US-based corporation Nvidia, which produces the most advanced chips. Just five years ago, Chinese data centres, research laboratories, and tech giants operated almost exclusively on Nvidia hardware, whose market share in China was around 95 percent. When the US government prohibited the export of the most advanced chips to China, the company produced pared-down versions of its chips that adhered to US export regulations but were nonetheless useful for Chinese buyers.
Then in April 2025, Nvidia stated that the Trump administration had also blocked the sale of these less-powerful H200 chips to China. The corporation’s CEO Jensen Huang reported that his company’s Chinese market share subsequently collapsed entirely, adding “I can’t imagine any policymaker thinking that that’s a good idea”. For corporations like Nvidia, China is not just another market but rather the single largest and most profitable market for AI chips, which have been driving the growth that sustains and justifies the tech industry’s staggering market valuations.
However, when the Trump administration once again granted export licences for certain Nvidia and AMD chips last August, something happened that it had not anticipated: suddenly the situation reversed and now it was Beijing that was prohibiting these chips from entering China. So while the US government was focusing on export permissions for Nvidia so that the company could maintain its position as market leader, the Chinese government was working to reduce its dependence on US technology. Chinese corporations had already developed high-performance chips that now, unburdened by foreign competition, could be scaled up to fill the gap in the Chinese market.
The key question is still whether the Chinese chips are of sufficient quality to replace the American ones. In the West, the idea predominates that Chinese semiconductors are cheap imitations, five to ten years behind cutting-edge technology from the US. That was actually true for a long time, but the gap appears to have shrunk recently. Huawei has developed the Ascend 910B chip, which is seen as being comparable to the Nvidia A100. It is not even the newest generation of chips, but it is still good enough for most AI training models. Alibaba developed the Yitian 710, a server processor for cloud computing, and has already rolled it out across the entirety of the company’s cloud infrastructure.
The New Sputnik Shock
Although Chinese AI developers still have ground to make up in terms of both chip design and, in particular, production, that is not necessarily a disadvantage when it comes to developing AI models, given that Chinese companies tend to make adversity a virtue when looking for innovative solutions.
When the Chinese company DeepSeek released its AI model in January 2025, it sent shockwaves through the tech and finance worlds due to its extremely low development cost of US$ 5–6 million — a vanishingly small sum in comparison with the multi-million dollar investments made by tech giants in the US like OpenAI or Meta. By using new techniques that revolutionized digital signal processing and data compression, DeepSeek was able to massively reduce the number of high-performance chips needed to run AI systems. The fact that the DeepSeek models were able to provide the same performance as AI models from Silicon Valley while using fewer resources shook people’s faith in the technological superiority of the US AI industry and led to the widespread adoption of the Chinese start-up’s product. Venture capitalist Marc Andreessen described it as a “Sputnik moment” for the US tech industry.
One year on, several Chinese AI models have established themselves worldwide. Alongside DeepSeek, Qwen and Kimi are among the models that have risen to the top of the rankings. These models are based on home-grown components: Chinese hardware and software; chips from Huawei, Cambricon, and the newcomer Moore Threads; and at the proprietary programme level, complex infrastructure and a full technological stack. This puts them at the heart of China’s algorithmic sovereignty.
Competition or Continental Drift?
In some areas, this has meant competition between the US and China for the best AI models and the highest-performing chips. Along the way, the gaps between US and Chinese companies are closing much more quickly than even recent forecasts had predicted. Additionally, there is the race for artificial general intelligence (AGI), which is aimed at emulating the cognitive abilities of the human brain and independently advancing these abilities. However, if the experts are correct, the industry remains far from reaching that goal for the time being.
Although this race is very real, the coming years may be defined primarily by the US and China’s respective approaches drifting further from one another. For one thing, the Chinese technology model seems to focus primarily on narrow, practical applications. Artificial intelligence is ubiquitous in everyday life in China: how can AI be used to improve public administration, education, and healthcare? How can AI be interwoven with robotics, and how can it help increase agricultural efficiency and industrial production? US corporations emphasize the development of the highest-performing models; it is up to other companies to then figure out for themselves how those models can be used for different applications. By contrast, China concentrates on the application and is pursuing a national strategy for introducing AI into its entire economy that way.
A second way in which China’s approach is distinct is that almost all AI models produced there (an estimated 85–90 percent) are open source. That is due to guidelines established by the Chinese state that force market participants to use open source models. In its promotion of the chip industry and AI sector, Beijing is essentially following the same playbook that it has already used in other industries like electric vehicles, green technology, batteries, and 6G telecommunications.
In order to implement a new technology, it has to be available in abundance, and it has to be cheap enough in comparison with older technologies that the financial costs of changing course are kept to a minimum. In this respect, the Chinese state is banking less on maximizing profit in the short term and more on maintaining its focus on the overarching strategic goal of quickly and extensively rolling out economic innovations. The Chinese open source model makes it easier to adapt to specific needs and local requirements not only in China itself, but also in other regions around the world. Moreover, Chinese hardware is 30–50 percent cheaper than its US competitors.
In that sense, technological trends seem to be in lockstep with political ones. That could result in a division in technological development and the establishment of separate, partially incompatible ecosystems: one AI system for the Global North, in which the world’s largest corporations, making use of their sizeable financial resources, opt for US models; and another in which the companies of the Global South, in search of cheaper alternatives to US technology, rely on Chinese products.
As an economic division of labour, that absolutely could work. However, in terms of geopolitical rivalry, this development could also fuel a new cold war.
Translation by Joseph Keady & Rowan Coupland for Gegensatz Translation Collective.
