The rivalry is not just technological – it will also spill over into broader geoeconomics
- The release of DeepSeek-R1 in January 2025 was, for many observers, a demonstration that hardware disadvantage need not mean capability disadvantage. IMAGE: REUTERS
FEW rivalries in modern history have carried as much consequence as the one now unfolding between the US and China over artificial intelligence.
Unlike previous technological competitions – the space race and the nuclear arms build-up – this one is not confined to a single domain.
AI is a general-purpose technology: one that involves semiconductors, scientific research, military affairs, global markets and the everyday lives of billions.
Whoever leads in AI does not merely win a trophy. They shape the infrastructure through which the 21st century will be organised.
Two visions, one prize
The competition between Washington and Beijing is not simply a race to build the fastest model. It is also a contest between two fundamentally different visions of how transformative technology should be developed and deployed.
The American approach has been characterised by private-sector dominance, concentrated investment in compute-intensive frontier models and a bet that superior hardware will ultimately yield superior capabilities.
AI investment in the US stands at roughly US$67 billion, compared with China’s US$44 billion. American firms are also raising private capital at around 2.4 times the rate of their Chinese counterparts.
OpenAI, Anthropic, Google and Meta form a formidable constellation of frontier labs, each racing towards capabilities that would have seemed fantastical a decade ago.
China’s path is different in character, if not in ambition. Constrained by US export controls on advanced semiconductors, Beijing has leaned into efficiency, openness and state-directed coordination.
The release of DeepSeek-R1 in January 2025 was a clarifying moment: a model that rivalled American systems at a fraction of the training cost, developed under resource constraints that most Western researchers assumed would be crippling.
It was, for many observers, a Sputnik moment in reverse – a demonstration that hardware disadvantage need not mean capability disadvantage.
Unintended consequences of curbs
Central to the competition is the struggle over semiconductors.
American policymakers, convinced that advanced chips are the critical chokepoint in AI development, have pursued an aggressive export control regime designed to deny China access to the most powerful semiconductors.
Before 2023, Nvidia held a 90 per cent or more share of China’s AI chip market. That dominance has since crumbled. Chinese domestic chips now account for nearly 41 per cent of the local market, with Huawei’s Ascend series leading the charge.
At the single-chip level, Chinese processors remain meaningfully behind Nvidia’s Blackwell on key metrics such as processing performance and memory bandwidth.
But the gap is narrowing, and the restrictions designed to maintain it may be accelerating China’s determination to close it.
Export controls, in other words, have both slowed China’s access to the frontier and turbocharged its motivation to reach it domestically.
A narrowing research gap
The US still holds an edge in research quality. American AI papers average a citation impact of 4.2, compared with 2.8 for Chinese publications.
But China leads decisively in volume, producing more than 41,000 AI papers annually against America’s 28,400. At the 2025 International Conference on Computer Vision, half of all the authors were affiliated with Chinese institutions, far exceeding the 17 per cent from the US.
In some of the most technically demanding corners of AI research, China is not catching up. It has already arrived.
Competing in the global market
The competition is not confined to laboratories and chip fabs. It is playing out in markets across the world.
Global large language model (LLM) usage trebled between April 2024 and August 2025, rising from roughly 2.4 billion to nearly 8.2 billion monthly site visits.
Chinese models have captured more than 10 per cent market penetration in 30 countries, and more than 20 per cent in 11.
Their advantage is partly economic. Chinese models are priced at one-sixth to one-fourth the cost of their American rivals.
It is also partly strategic, with Beijing pursuing an active AI diplomacy to embed its models in allied and developing nations’ infrastructure.
By releasing capable models freely, China is shaping the AI architecture of countries that might otherwise default to American providers. Several major US technology companies are already integrating Chinese LLMs into their own applications, an irony not lost on policymakers in Washington.
Military and security dimensions
AI is already accelerating a revolution in military affairs, in autonomous systems, cyberoperations, logistics and intelligence analysis.
The US government’s nearly US$11.2 billion arms package to Taiwan announced in late 2025 is a reminder that the AI competition is inseparable from broader geopolitical tensions over Taiwan, through which the world’s most advanced chips flow.
The congressional hearing on “China’s Campaign to Steal America’s AI Edge” in April 2026 underscored the security establishment’s view that this is not just a commercial rivalry, but also a strategic contest.
The testimony of researchers and officials at the hearing reflected a shared assessment: They alleged that China is pursuing a full-stack AI approach with the explicit goal of making its economy and military more capable across every dimension.
No clear victor
The honest answer to the question of who is winning is: It depends on what you are measuring, and no one knows yet.
The US retains clear advantages in frontier model capabilities, private capital and the quality of its research ecosystem.
It benefits from an extraordinary concentration of AI talent, much of it drawn from around the world. Its leading companies are pushing the boundaries of what AI systems can do faster than any comparable effort anywhere.
But China has demonstrated that it cannot simply be designed out of the game. Its research output is formidable.
Its domestic chip industry, though behind, is growing rapidly under the pressure of necessity. Its open-source strategy is winning market share across the Global South.
And its government’s willingness to treat AI as a strategic national priority gives it a coherence and long-term commitment that a market-driven ecosystem does not always provide.
A race without a finish line
Both countries would be unwise to assume the other is simply going to fall behind.
America should not mistake its current frontier advantage for permanent dominance, nor should it allow export control policy to become a substitute for the investment, education and regulatory clarity that a durable lead requires.
China, for its part, faces genuine constraints: in chips, in talent retention and in the economic headwinds that cloud its broader trajectory.
What is certain is that the decisions made in Washington and Beijing over the next several years will echo long past any particular model release or chip benchmark.
The AI race is, in the end, less a technical competition than a test of which society can most effectively translate technological capability into broadly shared human benefit. On that measure, neither country has yet made its case.