Beyond the Ban: China's Path to AI Supremacy
Market in Chinese city of Shenzhen, where many vendors sell banned microchips (Photo: The New York Times)
In the past decade, both the Trump and Biden administrations have expanded the US Department of Commerce’s Bureau of Industry and Security “Entity List,” imposing trade restrictions on individuals, companies, and organizations deemed national security risks. These moves, particularly aimed at limiting China’s access to advanced computing chips, reflect growing concerns over the intensifying "AI race" between the US and China. However, this race is not just a technological competition – it is increasingly straining international relations and undermining global efforts to regulate artificial intelligence.
At the second Responsible AI in the Military Domain (REAIM) conference hosted by Seoul in early September 2024, China opted-out of an international “blueprint for action” declaration committed to keeping AI out of nuclear authorization decisions. American officials are viewing the opt-out as Chinese protest over American export restrictions on foreign-made chips. However, these bans have proved counterproductive and ineffective, stifling American influence while failing to curb China’s AI advancements as their dominance becomes seemingly inevitable.
Since the Entity List only targets specific company names and addresses, Chinese firms have exploited a loophole by simply changing their names or locations to evade sanctions. After being added to the list in 2019, executives from Sugon, a supercomputer manufacturer linked to surveillance of ethnic minorities, quickly established Nettrix, now one of China's largest AI server producers. Both companies are tied to the Chinese Academy of Sciences, a key player in chip technology research, according to records accessed through WireScreen. Moreover, Sugon and Nettrix not only share a physical complex but also have overlapping executives, further blurring the lines between the sanctioned and unsanctioned entities.
Nettrix has established partnerships with US tech giants Nvidia, Intel, and Microsoft, selling servers equipped with American chips to organizations on the Entity List, effectively undermining the list’s intended purpose. By facilitating access to restricted technology through these partnerships, Nettrix has highlighted the limitations and loopholes in the export restrictions, calling into question the overall effectiveness of the US government's trade sanctions.
Nvidia's partnership with Nettrix underscores the reluctance of American businesses to fully curb China’s AI development, in stark contrast to the US government’s hardline stance. In late 2022, the US ordered Nvidia to halt shipments of its advanced A100 chips to China. Although Nvidia complied, it quickly introduced the minimally downgraded A800 chip, allowing sales to resume. US officials argued that the A800 enabled China to achieve nearly the same capabilities. In response to criticism from Washington, Nvidia’s General Counsel Tim Teter stressed that the new A800 was within the government parameters, likening it to driving 63 miles per hour with a speed limit of 65: “Am I violating the spirit of the rule? Of course not.”
Given that China is one of Nvidia’s largest markets, it’s unsurprising that the company doesn’t share the US government’s enthusiasm for restricting exports to China. In July 2023, Nvidia’s CEO, Jensen Huang, visited the White House alongside the heads of Intel and Qualcomm to argue that export controls were damaging to American businesses: cutting off access to such a significant market could weaken US companies, making them less competitive on the global stage. Since the meeting, Nvidia has announced a collaboration with Chinese entity Alibaba Group Holding’s cloud computing services unit. At the event announcing it on Sept. 20, 2024, Alibaba Cloud Chief Technology Officer Zhou Jingren highlighted the collaboration goals, saying that “together with our partners, we want to empower more businesses and individuals to unlock the potential of generative AI.”
Alibaba showcases plans and capabilities at Apsara Conference in Hangzhou (Photo: Xinhua)
While the US grapples with internal debates over AI export controls, China's rapid progress shows no signs of slowing down. In April of 2023, Alibaba introduced its open-sourced large language model, Tongyi Qianwen, which the company claims has been downloaded more than 40 million times. On Sept. 19, in its largest open-source initiative yet, Alibaba launched over 100 new open-source AI models, including a new text-to-video tool similar to OpenAI’s Sora, which generates video in response to a user’s prompt.
Chinese companies are embracing open-source AI, willingly sharing the underlying source code of their programs to accelerate growth of new artificial intelligence technologies. In contrast, while the White House has recently expressed support for open-source AI, the US has been facing internal disagreements that could hinder American open-source initiatives, inadvertently giving China a competitive advantage. As concerns about the potential for disinformation and defamation rise, particularly in this crucial election period (exemplified by candidates like Donald Trump leveraging AI to create fake endorsements) American companies and officials are increasingly wary. This trepidation complicates things, making it more challenging for the US to effectively harness the benefits of open-source AI while mitigating its risks.
As China makes rapid progress, certain American AI breakthroughs have come to depend on Chinese open-source technology. A Stanford University team attracted attention in June after they unveiled their AI model Llama 3-V, claiming it outperformed the existing leading models, only to discover that the model was built on an existing Chinese open-source system.
The Chinese government is supporting rapid AI development in their recent investments into its chip industry. May 2024 saw the nation set up its third state-backed investment fund for its domestic semiconductor industry, with investments from six of the country’s largest state-owned banks, including ICBC and China Construction Bank totalling roughly $47.5 billion. This marks a significant increase from the $19.2 billion and $28.2 billion allocated in the first two phases of the fund, in 2014 and 2019.
The United States' attempts to curb China’s AI development through export restrictions and sanctions appear increasingly futile, as Chinese companies find ways to circumvent trade bans and continue their rapid AI advancements. While American policymakers push for tighter controls, US businesses, eager to maintain their foothold in the Chinese market, resist fully cutting ties. Meanwhile, China's unified approach—bolstered by state-backed investment in its semiconductor industry and open-source AI initiatives—has propelled its AI sector forward, leaving the US struggling to maintain its technological edge. As China races ahead, the US must reconsider its strategy.