Beyond the Benchmarks: A Systemic View of U.S.-China AI Competition

Introduction

U.S.-China AI competition is one of the defining geopolitical dynamics of the 21st century—and the pace at which its stakes are escalating is itself part of the story. In our October 2025 report, The Geopolitics of AI, we surveyed the full geopolitical terrain being reshaped by AI, from Europe's pursuit of tech sovereignty to Middle Eastern capital flows and defense transformation. In this report, we focus on the competition that anchors all of them: the U.S.-China contest for AI leadership.

The release of Anthropic’s Mythos and similar models has offered a glimpse of the competition’s growing stakes. Mythos’s ability to identify cyber vulnerabilities that have evaded even the most advanced human experts and AI systems represents not only a significant advance in software assurance, but also cyber capability—one that could widen capability asymmetries overnight and push civilian systems across finance, energy, telecommunications, logistics, and health onto the front lines of interstate rivalry and the predations of non-state actors. China will develop comparable technology—and when it does, the competitive landscape will again shift.

As AI improves, the gap between a technical milestone and a geopolitical inflection point is closing faster than governments or companies are prepared for. At the same time, by over-indexing on fast-moving developments, decision-makers may underappreciate some of the AI competition’s slower moving dynamics. This report attempts to offer a framework that looks beyond snapshots of current capability to track leading indicators that reveal where this competition is heading.

The stakes of U.S.-China AI competition are high. AI’s potential to accelerate innovation and efficiency across technologies and sectors means that leadership here has the potential to translate into broader economic and firstmover advantages—and its applications across military and security domains make competition a national security imperative at a moment when the two countries are increasingly operating as peer competitors.

And those stakes are not lost on either government. Policymakers in Beijing and Washington have taken extraordinary steps over the last several years both to foster environments conducive to faster domestic AI innovation and adoption, and in some cases to widen the gap by restricting the other country's access to essential components of the AI supply chain. Yet despite a shared recognition of AI's strategic importance, the two governments are making fundamentally different bets about where that competition will be won.

• The U.S. strategy remains concentrated at the frontier. It is investing heavily in the most advanced hardware, backing the development of the world's leading models, and deploying abundant private capital in pursuit of raw capability leadership.
• China’s strategy is increasingly centered on deployment at scale. China is competitive at the model level, but its real emphasis is broader diffusion: rapid adoption at scale across the domestic economy, development of lower-cost and “good enough” open-source models, aggressive exports of Chinese technology stacks into global markets, and the data advantages, standards influence, and geopolitical leverage that come from being the infrastructure provider of choice across much of the developing world.

These are not just different resource allocations—they are different theories of how AI competition is won. The U.S. bet is that frontier capability translates into unassailable strategic advantage. China's bet is that adoption, scale, ecosystem embedment, and integration of AI into physical systems will ultimately prove more consequential. The two wagers are also manifestations of the two countries’ defining approaches: a U.S. commitment to exquisite systems, intangibles, and services; and a Chinese commitment to the practical, tangible, and manufactured.

Read our full report