US-China AI competition shifts as regulation clashes with China's cost-efficient model strategy

US Export Controls Create Friction in American AI Leadership

The US-China AI competition has taken an unexpected turn as the Trump administration's regulatory approach threatens to undermine American technological advantages. When Anthropic released its Fable 5 model in June, the US government swiftly issued export control directives blocking access by foreigners and noncitizens, forcing the company to disable the model entirely 2

2

. The government later permitted Anthropic to restore limited access to its Mythos model, though negotiations continue 2

2

. OpenAI CEO Sam Altman noted delicately that the restricted release process "isn't quite the process that we think is optimal" 1

1

. Reports suggest the administration has asked OpenAI to limit users for its next model as well 2

2

. This erratic export controls strategy has left the entire industry "frozen in place, waiting for something that seems kind of more coherent," according to former State Department technology adviser Daniel Remler 4

4

.

China's AI Progress Accelerates Through Open-Source Strategy

While US companies face mounting government control and private sector innovation in AI, China's AI progress continues through a fundamentally different approach. Zhipu AI's newly released GLM 5.2 model prompted David Sacks, co-chair of President Trump's Council of Advisors on Science and Technology, to declare it "as good as the currently available models from OpenAI and Anthropic" 1

1

. Chinese AI leaders increasingly see themselves running a different race, focused not on achieving technical benchmarks but on "how many people in the real world are actually using our models every day," said Cherie Shi, global business manager for MiniMax 5

5

. Top US companies sell Rolls-Royces while Chinese firms sell Mercedes, explained a Zhipu AI executive—American AI may be better for now, but Chinese products offer comparable capability at more affordable prices 5

5

.

Cost Efficiency Drives Global Adoption of Chinese Models

The global AI landscape is shifting as enterprises prioritize cost efficiency over cutting-edge capabilities. Companies reeling from skyrocketing AI costs are seeking cheaper alternatives, creating opportunities for Chinese firms that offer lower token rates and more efficient operations 5

5

. "If we can provide 80 percent of the value at a much lower cost, that's enough. We'll take that market share," said a salesperson at a leading Chinese AI firm 5

5

. Chinese AI firms are gaining significant ground across Southeast Asia and the Persian Gulf region. Singapore's government-backed AI initiative chose Alibaba's Qwen, while Saudi Arabia is partnering with ByteDance and Huawei for urban infrastructure projects 5

5

. Zhipu AI, restricted by the US Commerce Department for alleged military ties, has opened offices in Singapore and Dubai and is securing government contracts across Asia 5

5

.

Open-Source AI Models Challenge Closed Systems

The debate over open-source AI models has become central to the geopolitical AI race. Nvidia is developing open-weight models where anyone can inspect and adjust the mathematical weights, with CEO Jensen Huang betting that commodity AI will drive chip sales 3

3

. Palantir announced it will run Nvidia's Nemotron models inside classified government systems, with CEO Alex Karp arguing it's riskier to hand data to third parties than to secure it internally 3

3

. Emily Weinstein from the US-China Economic and Security Review Commission warns that widespread adoption of Chinese AI in the Global South could create a "Huawei model on steroids" where countries become reliant on Chinese infrastructure that isn't interoperable with US systems 4

4

.

AI Arms Control Faces Fundamental Obstacles

The possibility of multilateral AI arms control between the US and China remains uncertain. The Trump administration appears interested in talks with Beijing, though China has not agreed to formal negotiations 1

1

. Cyber security presents an immediate obstacle—both Chinese hackers and the US National Security Agency already use leading Anthropic models for operations 1

1

. Biosecurity controls face similar challenges, with China's transparency record raising concerns reminiscent of the 1972 Biological Weapons Convention when the Soviets expanded bioweapons work despite treaty commitments 1

1

. Discussions on autonomous weapons encounter practical barriers—no major power is likely to renounce automatic missile-defense systems 1

1

. Verification presents the biggest hurdle. Neither country will grant access to sensitive source code, making compliance assessment nearly impossible 1

1

.

Digital Sovereignty Challenges US Alliance Strategy

The State Department expanded its Pax Silica initiative this week to build a US-led AI and chip supply-chain bloc, with 35 countries signing the Declaration on AI Opportunity 4

4

. Undersecretary Jacob Helberg criticized digital sovereignty as "backward and counterproductive" and "synchronized mediocrity" 4

4

. Yet US partners are walking a tightrope. The European Union emphasized digital sovereignty following the Anthropic decision, while the UAE's assistant foreign minister told Axios their goal is achieving "strategic autonomy through international collaboration with trusted partners" 4

4

. Former Commerce Department counselor Saif Khan predicts "much of the rest of the world will likely, at least on the margin, prefer Chinese open-weight models" over US frontier AI 4

4

. The challenge for the US extends beyond maintaining technical leadership to convincing the world to keep building on American AI 4

4

.

Government Control Threatens Innovation on Both Sides

Both countries struggle to determine whether their frontier AI companies represent national champions or national security threats 2

2

. The Trump administration has careened between extremes, initially de-emphasizing safety concerns before designating Anthropic a supply chain risk in March after the company protested use of its models in autonomous weapons and domestic surveillance 2

2

. China requires security assessments and tests models for political sensitivities, has blocked companies from purchasing advanced Nvidia chips despite US approval, and reportedly restricted overseas travel by AI researchers 2

2

. The effects of Beijing's 2020 regulatory storm against tech companies remain evident 2

2

. Self-destructive American actions, not Chinese competition, may pose the most significant threat to advancements in AI for years to come 2

2

. Biden administration chip restrictions gave the US a compute advantage that kept American companies ahead in capability and revenue, but failed to prevent Chinese competitors from releasing high-quality models 1

1

. If the US tightly regulates domestic companies while Chinese firms operate more freely, it risks kneecapping American AI leaders without addressing global security concerns 1

1

. The right answer is more competition, not picking winners and losers through AI regulation that would hurt American startups while foreign competitors remain free to adopt cheap Chinese alternatives 3

3

.