Experts say the U.S. and China aren't in an AI race because they aren't targeting the same goals
President Donald Trump hosted Chinese President Xi Jinping for a state visit that touched on AI.
As Chinese President Xi Jinping made his first state visit to the U.S. in 11 years to meet with President Donald Trump, questions swirled over how the leaders would address one of the key competitive issues between the two countries: artificial intelligence.
Recent calls from leading U.S. AI labs Anthropic and OpenAI to slow the pace of AI development amid mounting safety fears were met with concerns that the approach would allow China to catch up in a race that has so far been dominated by the U.S.
However, some experts say viewing the competition as a winnable race misses the point.
"If the benchmark is the single strongest frontier model, the U.S. has an obvious advantage," said Lizzi Lee, a fellow on the Chinese economy at the Asia Society Policy Institute's Center for China Analysis, referring to cutting-edge AI systems.
However, "if the question is what model developers around the world can cheaply download, modify and deploy, China looks much stronger."
The differences in how AI has been developed and deployed within each country and then to other parts of the globe make a clear finish line basically nonexistent.
Still, the competition for supremacy will continue to shape the AI ecosystem for the rest of the world, experts told MarketWatch.
Two countries, two goals
Since the launch of OpenAI's ChatGPT in November 2022, the U.S. and China have developed distinct AI ecosystems.
U.S. labs at the forefront of AI development "have been operating in a world of abundant capital and compute," Lee told MarketWatch, referring to computing power offered by advanced chips, such as those from Nvidia (NVDA). Therefore, "the instinct has been to keep scaling" and pushing the boundaries of what is possible for AI to do, she said.
China, on the other hand, has been building its AI capabilities with a scarcity of advanced chips, making efficiency "a competitive advantage" for the country, Lee said in emailed comments.
The focus on open-weight models, or those whose parameters are public and available to be run and adjusted by other developers, makes sense commercially, she said. Chinese companies also "care a lot about whether models can actually run cheaply at scale."
By using open-weight models, developers can piggyback off of existing efforts to develop AI without having to do as much hardware-intensive training work.
China's open-weight offerings have been a way for the country to undercut U.S. companies.
It's "a way of promoting Chinese soft power internationally, promoting Chinese diplomatic objectives by signaling that they stand for broad access to these models in contrast to the Americans who are focused on restrictions and closed-source models," Sam Winter-Levy, a senior fellow in technology and international affairs at the Carnegie Endowment for International Peace said.
The split in approaches predated ChatGPT.
Early Chinese AI was considered a part of neuroscience, William Hannas, a lead analyst at Georgetown University's Center for Security and Emerging Technology, told MarketWatch. As late as the mid-2010s, China was conducting AI research with neuroscience funding, Hannas said - something he saw less of in the U.S.
China has also put more emphasis on the practical implementations of AI, Hannas said, pointing to the Chinese government's AI Plus strategy aimed at integrating AI across the country's economy and society.
The computing edge
Computing power is a key input for AI no matter where the competition is happening, Winter-Levy told MarketWatch.
Training and running frontier-level AI models requires vast quantities of computing power, but so does deploying AI across an economy that scales to millions of people, he noted.
"We're seeing right now that the Chinese companies face real constraints when it comes to access to compute, not just for training models, but also for deploying them," Winter-Levy said.
Often when a Chinese company has released a new model, it has had difficulty serving it to large numbers of people, resulting in outages and delays, he added.
Although U.S. export controls restricting China from accessing advanced chips have been a constraint, Beijing has pushed homegrown labs to find alternatives around that. DeepSeek was said to have developed competitive AI technology with limited chip supply. This suggested that having large amounts of advanced computing power - as is the case for leading U.S. labs - doesn't necessarily translate into more capable AI models, Lee said.
Export controls can also be essentially irrelevant in some cases, Winter-Levy said, such as for training smaller models like those for biosecurity or missile systems, which can rely on older chips.
Still, computing power remains an advantage for the U.S. and a major principle of both U.S. and Chinese policy on AI, Lee said. While the U.S. has focused on slowing China's ability to develop cutting-edge AI models, China views chip-export restrictions as a reason why it can't rely on a U.S.-controlled technology stack.
"Each side's response reinforces the other side's threat perception," Lee said, and that loop remains hard to manage.
Regulating different priorities
Government involvement in AI regulation has become a hot-button issue in the U.S. over recent months. People both inside and outside the industry have become concerned about reports of autonomous AI agents breaking out of containment and hacking other companies.
In China, however, regulation was an early focus for the government, according to Lee. That ties in with its emphasis on controlling information, securing data and maintaining social stability.
For now, China generally seems less alarmed by the types of risks U.S. frontier labs are warning about, Winter-Levy said, but that is partly due to Chinese companies running behind.
"If we start to see that happening in China with Chinese models too, then the Chinese may take those risks more seriously," he said.
Don't miss: The murky AI milestone that has some of the industry's leading voices increasingly on edge
-Britney Nguyen
Comments