US-China AI Hegemony Rivalry Intensifies: Moving Beyond Technology to Regulatory Warfare

The artificial intelligence hegemony race between the United States and China is expanding beyond technological development into a competition over regulations and market share.

Concerns are growing in the U.S. as "Kimi K3," the latest model recently unveiled by Chinese AI startup Moonshot AI, has been evaluated in some performance assessments to match or even surpass the latest models from OpenAI and Anthropic.

The U.S. government suspects that Chinese companies are engaging in "distillation," utilizing the output generated by existing AI as training data to develop high-performance models at a low cost.

Axios reported that the U.S. government is considering measures to effectively block domestic companies from accessing Chinese AI products through government procurement regulations or by adding them to trade restriction lists.

U.S. Treasury Secretary Scott Bessent also said in a Fox Business interview on the 21st that Chinese models sent "hundreds of millions of calls to U.S. models," adding, "This is unacceptable, and we will thoroughly investigate this matter within days or weeks."

China has also taken counteraction.

China's Ministry of Commerce is reportedly considering measures to restrict foreign companies' access to core AI data and the acquisition of major tech startups.

Discussions are also underway to restrict the overseas transfer of core data used for AI training while allowing overseas users to download models and use services.

Previously, Reuters reported that China is considering a "three-tier rating system" that would bundle basic open-source models with simple notifications, advanced technologies with security reviews, and frontier-grade models exclusively for domestic use.

China is also strongly protesting U.S. suspicions regarding "distillation."

China's Ministry of Foreign Affairs dismissed the related claims as baseless accusations, arguing that certain countries are exaggerating the issue.

Amid ongoing government-to-government disputes, corporate choices are moving in a more pragmatic direction.

While companies previously prioritized using the latest and highest-performing AI models, a so-called "model maxing" strategy—sharing and dividing models according to the nature of tasks—is spreading recently.

This approach involves using high-performance U.S. models for complex reasoning or research tasks, while utilizing cheaper Chinese or open-source models for simple repetitive tasks.

In fact, according to data compiled by AI model marketplace OpenRouter, the share of Chinese models among the total token usage by U.S. companies reached 63% as of early this month.

This is a sharp increase from less than 10% a year ago.

In particular, Chinese models such as DeepSeek, Qwen, Kimi, and GLM accounted for nearly half of total usage, and the usage share of the DeepSeek single model reportedly surpassed Google, OpenAI, and Anthropic combined.

Analysts point out that price competitiveness lies behind this widespread adoption.

According to Citi Research, the cost of using Chinese models averages about 18 cents per 1 million tokens, whereas top-tier U.S. models average around $4.

Flo Crivello, CEO of AI startup Lindy, emphasized cost efficiency, stating, "You don't need top-performing AI to write a single email."

Coinbase CEO Brian Armstrong also projected that within one to a year and a half, 80% of AI tasks will be handled by much cheaper models than today.

However, debate surrounding the proliferation of Chinese open models is also intensifying within the AI industry.

Anthropic CEO Dario Amodei warned in a recent Bloomberg interview that without regulation, it could lead to a "dystopian" AI future, noting that if powerful AI is provided free to anyone, it could pose serious risks in areas such as cybersecurity.

On the other hand, some investors disagree with viewing the spread of Chinese open models as a security threat, arguing it is merely the result of market competition.

Venture capitalist Bill Gurley argued in an op-ed for The Washington Post, "Lobbyists are pushing Washington to treat open models as a security threat. But the reality is normal competition that should be welcomed."

This debate recently spilled over into a public war of words between U.S. government officials and OpenAI executives, leading analysts to suggest that the U.S.-China AI competition is expanding beyond technology and regulation into a struggle for dominance across the entire industry.
※ Please note: This article was translated by AI and may contain errors.