CSET Says US and China Are Pursuing Different Paths to AGI

Two definitions of the AGI challenge

The Center for Security and Emerging Technology, or CSET, said the United States’ frontier-model ecosystem generally defines AGI around broadly human-level intellectual abilities across many cognitive tasks. This view supports investment in larger models, advanced reasoning and agents that can complete extended sequences of work.

Chinese researchers also develop competitive large language models, including open-weight systems. However, prominent Chinese discussions of 通用人工智能, which literally translates to “general artificial intelligence,” place greater emphasis on intelligence grounded in interaction with the world.

That difference helps explain China’s sustained interest in embodied AI, brain-inspired computing and combinations of human and machine cognition. CSET said the central unresolved question is whether scaling current model architectures will be enough, or whether major advances will require learning through action, biological inspiration or hybrid systems.

Where each country has an advantage

CSET identifies an American edge in semiconductors, hyperscale computing, frontier reasoning models, private investment and influential research institutions. The United States also benefits from global content and sophisticated cloud ecosystems, while its ability to attract global experts supports its research position.

China’s strengths include formidable engineering capacity, a large technical workforce, expanding electrical infrastructure and the ability to deploy AI in physical environments at scale. It has a vast STEM pipeline and is increasingly able to retain domestic talent while reversing its earlier “brain drain.”

China is responding to its relative weakness in frontier computing through domestic substitution and engineering techniques that make constrained hardware more productive. According to CSET, these efficiency gains narrow the performance gap without removing the U.S. hardware advantage. China also benefits from large-scale industrial deployment and coordinated data initiatives, although censorship and data fragmentation create limitations.

Why robotics and neuroscience are part of the race

If general intelligence requires more than processing text and other digital inputs, the physical world could become an important training environment. Embodied AI systems perceive, act and learn through interaction, potentially drawing on China’s strengths in robotics, manufacturing, electric vehicles and industrial automation.

CSET cautions that this does not establish a Chinese lead. American companies are also advancing robotic cognition, making embodiment an emerging area of competition. China additionally views neuroscience as a possible source of breakthroughs through brain-inspired architectures, neuromorphic hardware and brain-computer interfaces.

It remains unknown whether these approaches will outperform contemporary AI methods or complement them. CSET says their strategic importance partly comes from providing alternatives if current methods encounter limits.

What CSET says policymakers should watch

The analysis says current model rankings should not be treated as a permanent strategic verdict. Neither country’s portfolio guarantees AGI, and the path to AGI or artificial superintelligence remains uncertain.

CSET recommends strengthening the foundations of American leadership, including advanced chips, private capital, electricity, grid capacity, research talent and the ability to turn breakthroughs into practical systems. It also calls for avoiding a single-track strategy by supporting frontier models while exploring embodiment, neuroscience-inspired methods and architectures that could complement or displace current approaches.

The forthcoming report argues for selective protection of high-risk capabilities alongside narrowly bounded cooperation where mutual interests are compelling. It also urges sustained monitoring of China’s industrial systems, deployments, alternative architectures and concepts of AGI, rather than relying only on headline benchmark scores.