IIn 2017, the Chinese government announced an ambitious roadmap to lead artificial intelligence development, aiming to secure global leadership by 2030. The plan calls for “symbolic advances” in AI to be demonstrated and demonstrated by 2020. Then, in late 2022, OpenAI’s release of ChatGPT surprised the world and caught China off guard.
At the time, China’s big technology companies were still reeling from an 18-month government crackdown that had shaved about $1 trillion from China’s technology sector. It took almost a year for a handful of Chinese AI chatbots to receive government approval for public release. Some wondered whether China’s stance on censorship would hinder the country’s AI ambitions. Meanwhile, the Biden administration’s export controls, announced just a month before ChatGPT’s debut, were aimed at cutting China off from advanced semiconductors essential to training large-scale AI models. Without cutting-edge chips, Beijing’s goal of achieving AI supremacy by 2030 looked increasingly unattainable.
But fast forward to today, and China’s impressive string of releases suggests that America’s AI lead is shrinking. In November, Alibaba and Chinese AI developer DeepSeek released an inference model that is in some ways comparable to OpenAI’s o1-preview. That same month, Chinese video game giant Tencent announced its open source model Hunyuan-Large, which outperformed top U.S.-developed open source models on several benchmarks in its tests. And in the last days of 2024, DeepSeek released DeepSeek-v3. It currently ranks highest among open source AIs on popular online leaderboards, and holds that position compared to top-performing closed systems from OpenAI and Anthropic.
Read more: How the benefits and harms of AI will grow in 2024
Before DeepSeek-v3 was released, the trend had already caught the attention of Eric Schmidt, former CEO of Google and one of the most influential voices on U.S. AI policy. In May 2024, Schmidt confidently claimed that the United States had a two- to three-year lead in AI, saying, “In my opinion, this is forever.” But by November, speaking at Harvard Kennedy School, Schmidt changed his tune. He cited the progress of Alibaba and Tencent as evidence that China is closing the gap. “This is shocking to me,” he said. “I thought the restrictions we put on chips would prevent them from circulating.”
Beyond sources of national prestige, who takes the lead in AI could impact the global balance of power. If AI agents can automate large portions of the workforce, they could boost national economies. And future systems that can direct weapons or hack enemies could provide a decisive military advantage. Artificial intelligence could emerge as a powerful tool for global influence as countries caught between these two superpowers are forced to choose between Chinese and American AI systems. China’s rapid progress has raised questions about whether U.S. semiconductor export controls are sufficient to maintain U.S. dominance.
Read more: How Israel is using AI in Gaza — and what it means for the future of the war
Building more powerful AI relies on three key ingredients: data, innovative algorithms, and raw computing power: compute. Training data for large language models like GPT-4o is typically scrapped from the internet, making it available to developers around the world. Similarly, new ideas for improving algorithms, or AI systems, can easily cross national borders, as new techniques are often shared in academic papers. Even if it doesn’t, China has a wealth of AI talent, producing more top AI researchers than the United States. In contrast, advanced chips are incredibly difficult to manufacture and, unlike algorithms or data, are physical products that can be stopped once and for all. border.
The supply chain for advanced semiconductors is dominated by the United States and its allies. US companies Nvidia and AMD have a de facto duopoly in data center GPUs used for AI. Their designs are so complex, and transistors are measured in single-digit nanometers, that Taiwan’s TSMC is currently the only company making these top-of-the-line chips. To do so, TSMC relies on a multi-million dollar machine that can only be manufactured by Dutch company ASML.
The US is trying to use this to its advantage. In 2022, the Biden administration introduced export control legislation that would prevent the sale of cutting-edge chips to China. The move follows a series of measures begun under Trump’s first administration to limit access to China’s semiconductor manufacturing technology. These efforts not only limit the flow of advanced chips into China, but also hinder the domestic chip industry. U.S. Commerce Secretary Gina Raimondo said on 60 Minutes in April that Chinese chips are “years behind.”
Read more: Study finds stark global disparity in ownership of powerful AI chips
But the 2022 export ban faced its first hurdle before it was announced, with Chinese developers reportedly stockpiling soon-to-be-restricted chips. DeepSeek, the Chinese developer behind an AI inference model called R1 that rivals OpenAI’s O1 preview, shipped a cluster of 10,000 soon-to-be-banned Nvidia A100 GPUs a year before export controls were introduced. I assembled it.
Smuggling may also have undermined the effectiveness of export controls. In October, Reuters reported that restricted TSMC chips were found in products made by Chinese company Huawei. It has also been reported that Chinese companies have used shell companies outside China to obtain restricted chips. Some companies are circumventing export controls by renting GPU access from offshore cloud providers. In December, the Wall Street Journal reported that the United States was preparing new measures to restrict China’s access to chips through other countries.
Read more: Has AI progress really slowed down?
U.S. export controls limit China’s access to cutting-edge semiconductors, but still allow it to sell inferior chips. Determining which chips should and should not be allowed has proven difficult. In 2022, Nvidia tweaked the design of its flagship chip to create a version for the Chinese market that fits within its limits. The chip remains useful for AI development, and the U.S. will tighten regulations in October 2023. “There was a year when[China]was able to buy chips of basically the same quality,” said AI and computing leader Lennart Heim. RAND Corporation Technology and Security Policy Center. He said export controls have yet to fully impact China’s AI development because of this loophole and the time it takes for new chips to be introduced into AI developers’ infrastructure.
It remains to be seen whether the current thresholds strike the right balance. In November, Tencent released a language model called Hunyuan-Large that outperformed Meta’s most powerful Llama 3.1 variant on several benchmarks. Although benchmarks are an imperfect measure of comparing the overall intelligence of AI models, Berkeley Risk and Security Lab research shows that Hunyuan-Large’s performance is better than that of the less powerful and unrestricted Nvidia H20 GPU. This is impressive because it was trained using the . “It’s clear that hardware is being better utilized because of better software,” said study author Ritwick Gupta, an advisor to the Pentagon’s Defense Innovation Unit. Rival China Lab’s DeepSeek-v3 is considered the most powerful open model available, but it was also trained using surprisingly little compute. Although there is great uncertainty about how President-elect Donald Trump will approach AI policy, multiple experts told Time magazine in November that they expect export controls to remain in place or even expand. He said he was doing it.
Before new regulations were introduced in December, Chinese companies again stockpiled soon-to-be-blocked chips. “We need to rethink this whole strategy,” says Mr. Gupta. “Stop playing whack-a-mole with these hardware chips.” He said the U.S. should not try to slow down the development of large-scale language models by restricting access to the chips, but instead suggests that we should focus on preventing the development of artificial intelligence systems. Developing military AI systems often requires less computing power for training, he said. But he acknowledged that restrictions on other parts of the chip supply chain, such as ASML’s machines used to make chips, have been critical to slowing China’s domestic chip industry.
Heim says the U.S. gap has narrowed in the past year, but China may now match the best U.S. open source models, but is about a year behind the top closed models. points out. He added that a narrowing gap does not necessarily mean export controls are failing. “Let’s move away from the binary of whether export controls are working or not,” he said, adding that it may take longer for China to feel that export controls are hurting.
The past decade has seen a dizzying increase in the amount of computing used to train AI models. For example, OpenAI’s GPT-4, released in 2023, is estimated to have been trained using approximately 10,000 times more compute than GPT-2, released in 2019. There are signs that American companies such as X and Amazon will continue this trend. Build massive supercomputers with hundreds of thousands of GPUs, far exceeding the computational power used to train today’s leading AI models. If that happens, Heim predicts that U.S. chip export restrictions will prevent China from keeping up in AI development. “Export restrictions mainly affect volume,” Heim said, noting that even if some of the restricted chips end up in the hands of Chinese developers, export restrictions will reduce the number, so models It added that it would be difficult to train and deploy on a large scale. “Unless the importance of computing changes, we generally expect the impact of export controls to become more severe over time,” he says.
Within the U.S. government, “there’s a reluctance right now to bring China to the[negotiating]table,” said Scott Singer, a visiting fellow in the Technology and International Affairs Program at the Carnegie Endowment for International Peace. Implicit reasoning: “(If the US is in the lead) why should we share anything? ”
But he points out that there are compelling reasons to negotiate with China on AI. “China doesn’t necessarily have to be a source of catastrophic risk,” he said, noting that China’s continued progress despite its computing limitations could one day produce AI with dangerous capabilities. He added that it means there is. “If China is closer, consider what kind of conversations you want to have with China to ensure the security of both systems,” Singer said.
