A Thor Dynamics analysis finds China has more AI papers at ICML 2026, 2,514 points to 1,849 for the U.S., while the U.S. keeps its edge in frontier models. Carnegie Mellon's Griffin-funded Miami campus plans to enroll students in 2028.
China has more AI papers at ICML 2026, according to a new analysis by Thor Dynamics. China scored 2,514 points. The United States scored 1,849.
Thor Dynamics is a defense and space photonics company. Its CEO, Gleb Chuvpilo, wrote the analysis, which was published on Medium.
How Wide Is the Gap?
Each accepted paper earns one point, split among its listed institutions. China holds 37.9% of the total score. The U.S. holds 27.9%.
Thor Dynamics ran the same test eight years ago on NIPS 2017, now called NeurIPS. The U.S. led that year, 414 to 39, or 10.6 to 1.
The ratio now favors China, 1.36 to 1. Thor cautions that the two conferences differ, so the swing is not an exact measure. It calls the result “a very large signal.”
Which Institutions Lead?
Tsinghua University ranks first with 165.2 points. The Chinese Academy of Sciences, Shanghai Jiao Tong, Zhejiang and Peking universities follow.
MIT is seventh at 89.7. Google is eighth at 83.0. Carnegie Mellon University is ninth at 79.2.
Seven of the top ten are Chinese or Hong Kong institutions. Thor counts Hong Kong with China.
Carnegie Mellon ranked second in the 2017 list, behind Google. That comparison spans different conferences and scoring methods.
Why Did China Move So Fast?
Thor points to several causes at once. China released its Next Generation Artificial Intelligence Development Plan in 2017, aiming to lead AI innovation by 2030.
Universities built machine-learning programs, trained more graduate students and sent them into new labs. Thor describes a self-reinforcing loop.
Industry joined in. Alibaba, Tencent, Huawei, ByteDance, Baidu and Ant now publish internationally competitive research, Thor says.
The field also grew. Thor notes that top conferences are far larger than a decade ago, so China did not displace American researchers. “The stadium got much bigger.”
Other countries trail far behind. South Korea scored 291, the United Kingdom 269 and Germany 220.
How Was It Measured?
Thor split each paper’s single point equally among the institutions listed on it. A paper listing MIT, Tsinghua and Google gives each 0.333.
About 3.4% of the score could not be assigned to a country. Thor says that residual does not change the headline.
Thor plans to repeat the test on NeurIPS 2026. It also wants to check whether China’s lead holds among oral and spotlight papers only.
Does Research Output Equal Frontier AI?
Thor says it does not. The metric does not measure compute, chips, funding, deployment or proprietary research.
On notable frontier models, the U.S. still leads, Thor says, citing Stanford’s AI Index.
Thor frames it as a split. “China may now have the world’s largest academic AI research engine, while America still possesses the world’s strongest frontier-model industrial engine.”
The costliest models need compute, infrastructure, proprietary data and capital. Thor says those resources increasingly sit inside companies, not universities.
Which advantage matters more over the next decade? Thor’s answer is short. “We don’t know.”
Where Does Miami’s Carnegie Mellon Gift Fit?
Ken Griffin committed $3 billion to Carnegie Mellon University on Sept. 30. Of that, $2 billion funds a new Miami campus, Fortune and USA Today report.
It will rise on a 35-acre site in Wynwood, subject to regulatory approvals. Construction is expected to begin in 2027, with first students in 2028 under the current plan.
At full build, it would support more than 3,500 undergraduate, master’s and PhD students, plus nearly 300 faculty.
Carnegie Mellon says CMU Miami will “organize education and research around major societal challenges rather than traditional academic disciplines.”
The stated focus areas are health and biological discovery, national security and strategic capabilities, and energy and climate resilience.
Griffin moved Citadel from Chicago to Miami in 2022. MAIN has covered Carnegie Mellon's $3B Griffin gift.
Laurence Ales, interim dean of CMU’s Tepper School of Business, says Miami has a greater concentration of information technology and finance. Both could bring faculty and students new partners.
Miami Dade College President Madeline Pumariega says the city must make sure “it anchors its talent.” Urbanist Richard Florida told Fortune that Griffin “realized that it has to be a talent anchor.”
Talent is also the engine in Thor’s data. China’s lead rests on a growing pool of professors, PhD students and research engineers.
Miami has research activity of its own. MAIN has covered University of Miami engineers teaching AI to admit uncertainty.
The Open Question
Thor’s data counts papers. CMU Miami describes its mission by problems: health, security and energy.
Publication volume and frontier capability are different measures, and Thor does not say which will shape the next decade.
Both engines now run at scale. Miami’s campus will open into that field in 2028.
Sources:
- "AI Research Rankings 2026: China Just Overtook America in AI Research," Gleb Chuvpilo, Thor Dynamics, Medium
- "What subject areas Carnegie Mellon Miami will focus on," Sarah Perkel, USA Today, Oct. 1, 2026
- "Two Cities. One CMU.," Carnegie Mellon University, September 30, 2026
- MAIN: Carnegie Mellon's $3B Griffin gift
- MAIN: UM engineers teach AI to say I'm not sure