China’s AI industry has drawn increasing media attention as its progress generates excitement and trepidation about a global future fueled by Chinese AI. One dimension of this success is the ability of Chinese actors, such as DeepSeek, to circumvent U.S. restrictions on the export of critical technology. According to Reuters, a U.S. official claimed this week that DeepSeek had evaded export controls to gain access to American AI chips. Earlier this month, The Wall Street Journal reported that Chinese engineers transported hard drives with hundreds of gigabytes of AI training data in suitcases to Malaysia in order to bypass U.S. restrictions by using American chips outside of China. But the flipside to this story is how U.S. export controls have encouraged the flourishing of China’s domestic AI ecosystem. Ann Cao and Wency Chen at the South China Morning Post described on Saturday how, following years of U.S. sanctions, Huawei’s advances in home-grown AI chips have allowed it to leapfrog some of its American competitors:
Jensen Huang, founder and CEO of AI chip giant Nvidia, has been the most prominent industry leader to recognise the resurgence of Huawei in the IC sector.
“All in all, the export controls were a failure. The facts would suggest it,” Huang told reporters on the sidelines of last month’s Computex expo in Taipei. He called on the White House to lower barriers to AI chip sales before American firms cede the China market to rivals like Huawei.
[…] The performance of [Huawei’s] Ascend chips against Nvidia’s in-demand GPUs was put under the spotlight this week, following the release of a technical paper that was jointly written by researchers from Huawei and Chinese AI infrastructure start-up SiliconFlow.
According to the paper, Huawei’s Ascend-powered advanced data centre architecture – CloudMatrix 384, along with the serving solution CloudMatrix-Infer – outperformed the Nvidia GPU-based SGLang fast-serving framework for large language models (LLMs), on both the inference and decoding phases, in running DeepSeek’s R1 reasoning model.
[…] “Huawei is a generation behind in chips, but its scale-up solution is arguably a generation ahead of Nvidia and AMD’s [Advanced Micro Devices] current products on the market,” the SemiAnalysis report said. [Source]
This month in Western media, experts have continuted to anxiously debate the extent of China’s progress in AI. In a Foreign Policy discussion on whether China can “catch up on AI,” George Lee of the Goldman Sachs Global Institute conceded that “we’re in the sprint mode of a real race for supremacy between the United States and China.” In a Foreign Affairs article titled “What if China Wins the AI Race?,” Sebastian Elbaum and Adam Segal argued that “the gap between U.S. and Chinese cutting-edge AI capabilities is narrowing, [and] American supremacy in AI is far from assured [….] Washington needs to plan for a possible future in which the United States loses the AI competition to China—or, at the very least, one in which Chinese AI models are as popular globally.” In a collection of commentaries at Brookings, Ryan Hass argued, “Rather than obsessing over which country is in the lead and what more the United States can do to slow China’s progress, U.S. policymakers must quickly gain comfort with the fact that America and China are going to be navigating the frontiers of AI side-by-side over the coming years.” In a ChinaFile conversation titled “Is China About to Produce the Next ‘Sputnik Moment’?,” Lizzi C. Lee highlighted how the industrialization of China’s AI ecosystem may soon propel it past the U.S.:
The sector to watch is AI. Chinese leaders increasingly recognize that control over AI depends on more than just models. It hinges on chips, energy, data centers, cooling systems, and even power grids.
Small Modular Reactors, or SMRs, in particular. They have long been viewed with cautious optimism in clean energy circles. In China, they are quickly becoming an emerging backbone of the country’s AI infrastructure strategy. The logic is simple: If compute (industry jargon for the infrastructure powering AI) is the engine of AI progress, energy is the fuel. And China wants control over the pump.
[…] So what might the “Sputnik moment” look like? Perhaps a frontier AI model, trained end-to-end in a nuclear-powered inland data hub, running on Huawei chips. Or a Belt and Road-style deployment of Chinese-built SMRs powering AI-driven logistics in Southeast Asia or Africa, proof that China can export not just textiles, steel, or iPhones, but the off-grid infrastructure needed to run the future of high tech.
The U.S. remains ahead in frontier innovation. But China is innovating in how to industrialize that frontier, specifically how to engineer cost-effective solutions for AI deployment and commercialization, harden the supply chain, and construct a full-stack ecosystem that is sanction-proof and self-sustaining. Quietly, it is building the scaffolding of its economy for the AI era. When that scaffolding becomes visible, that is when the real Sputnik moment will land. It will not arrive with spectacle, but it may arrive first! [Source]
Recent developments also demonstrate commercial innovation in China’s AI ecosystem. This month, Shanghai-based AI start-up MiniMax launched an open-source reasoning model that reportedly requires half the computing resources of DeepSeek-R1 and achieves similar performance to models by U.S.-based Antropic, OpenAI, and Google. Elsewhere, Chinese influencer Luo Yonghao and co-host Xiao Mu raked in over $7.65 million in six hours by using Baidu-AI-generated digital avatars to interact with viewers in real time on e-commerce livestreaming platform Youxuan, marking a potential “DeepSeek moment for China’s entire livestreaming and digital human industry,” Luo’s colleague boasted. Two months ago, China showcased its strides in embodied AI deployment by hosting the world’s first humanoid-robot half-marathon in Beijing, with mixed results.
The global repercussions of U.S.-China AI competition are coming into focus. On the hardware side, according to data analyzed by researchers at the University of Oxford, American and Chinese companies operate over 90 percent of the world’s data centers used for AI work, which has split the world into countries that rely on the U.S. and those that rely on China. Some African policymakers are working with Huawei to convert existing data centers to include Chinese-made chips, given U.S. restrictions on Africans’ access to American chips. On the software side, the global proliferation of Chinese AI services poses certain governance challenges for democratic societies that adopt them, as Kai-Shen Huang recently argued in The Diplomat. Chinese AI has been leveraged for surveillance, censorship, and propaganda within China, and DeepSeek and other Chinese AI models are already being instrumentalized by authoritarian actors abroad to extend their influence at the expense of democracy, as Alex Colville argued this week in Lingua Sinica:
[G]lobal access to an admittedly powerful — and, so far, free — AI model does not necessarily mean democratization of information. This much is already becoming clear. In fact, without proper safeguards, DeepSeek’s accessibility could transform it from a democratizing force into a vehicle for authoritarian influence.
Look no further than another country with big ambitions for AI development: India. Shortly after R1’s global launch Ola, an Indian tech giant, appeared to adapt and deploy a version of R1 to suit India’s information controls. It answered sensitive questions on China that the Chinese version refuses to discuss. But when questioned about anything critical of the government of Indian prime minister Narendra Modi, it refused in the same way the Chinese version would do about its own government: claiming the topic was beyond its abilities, and giving no answer.
Governments and experts have argued DeepSeek has few problems beyond small amounts of “half-baked censorship,” and data security issues. They must take it more seriously as a threat to freedom of expression. In our research at CMP, we have found that Chinese Communist Party bias is increasingly permeating the model with every new update, but tech companies are doing little (if anything) to retrain the model in ways that remove or otherwise temper these biases.
DeepSeek is indeed a boon for more accessible AI around the world, just as some have argued. But in the wrong hands, it also has the potential to be not just a vehicle for Chinese propaganda and information suppression, but a tool for authoritarianism worldwide. [Source]