Subscribe to The Informer for monthly expert analysis, and to Events for advance notice of visiting world leaders and distinguished guests.
You may unsubscribe from Lowy Institute newsletters at any time. For information on our privacy practices and how to unsubscribe, see our Privacy Policy.
Who is winning the AI race between the United States and China? The answer depends on what you measure.
Asia Power Snapshot: China, the United States, and the race for AI dominance
US President Donald Trump has declared that “whoever wins AI wins,” suggesting that the competition for mastery over artificial intelligence (AI) is the pre-eminent geopolitical contest of the day. But who is winning the AI race between the United States and China? The answer depends on what you measure. This Data Snapshot compares the inputs and outputs that matter most — from chips and compute to research and innovation — to show which side holds the lead, where the gaps are narrowing, and where capabilities may be overstated.
Based on a thorough analysis of available data, including 14 indicators across four measures — Resources, Technology, AI capabilities, and AI adoption — the United States is currently leading the race. However, its lead is far from uniform, and the competition remains uneven across multiple domains. The United States is well ahead in AI capabilities, driven by its dominance in compute, and in AI adoption, where its lead in web traffic and apps outweighs China’s edge in open models. By contrast, China has a clear lead in Resources, particularly critical minerals — the key materials needed for manufacturing the advanced technologies that make AI possible. Governance, however, remains contested. Neither superpower is widely trusted to regulate AI, leaving room for middle powers such as Australia to help shape the rules.
China holds a clear advantage over the United States in both production and known reserves of critical minerals.
In November 2025, the US Department of the Interior released a list of 60 minerals critical to its economic and national security interests. China leads in production of 30, while the United States is ahead in only eight — with 22 not produced in either country. China holds larger known reserves of 20 listed minerals, and the United States of seven, with one tied, and insufficient data for the remaining 32.
China’s ability to impose targeted export controls on critical minerals poses a major risk for the United States. In 2023, Beijing began announcing restrictions on several, including gallium and germanium — both used in semiconductors.
To counter this vulnerability, the United States has sought to cooperate with other countries to secure critical minerals. It has developed Pax Silica, a 24-member international framework focused on AI-related supply chain resilience. In Asia, it has critical minerals frameworks with Australia and Japan and has also signed memoranda of understanding with Malaysia and Thailand. In February 2026, it held a Critical Minerals Ministerial meeting, where it signed an additional 11 bilateral agreements, including one with the Philippines.
* For more on the semiconductor supply chain in Asia, and Southeast Asia’s role as a connector between the United States and China, see Between the superpowers: Southeast Asia’s strategic supply chain dilemma (Opens in new window).
The United States leads China in technological sophistication. While China has almost 11 million full-time equivalent researchers — over seven million more than the United States — US spending on research and development (R&D) is considerably higher as a share of GDP (3.45 per cent versus China’s 2.58 per cent). The United States also has a sizeable lead in the number of data centres and supercomputers.
The United States has sought to limit China’s access to advanced technologies that could help develop its semiconductor industry. Washington has imposed numerous restrictions on both China and Chinese subsidiaries abroad, which has, in turn, spurred affected companies to innovate around them. In May 2026, Huawei announced a new chip-design approach intended to increase transistor density without shrinking transistors. This could allow it to design chips by 2031 with transistor density equivalent to that of the chips Taiwan’s TSMC aims to make in 2028, without the advanced lithography machines China is barred from importing. If successful, this could limit the effect of US export controls.
While companies have looked for ways to work around US controls, Beijing has moved to expand its own, building on its critical mineral restrictions. In October 2026, China proposed adding 40 technologies to its restricted exports, including 23 related to AI — among them AI chip design, the core technology of the most advanced models, and AI tools, such as deepfake detection systems.
The United States maintains a lead in AI capabilities. Its AI data centres have nearly 40 times the combined computing power of China’s, and its AI chips 31 times. This has helped US models stay ahead in benchmark tests, though the margin is narrowing.
As of November 2026, the top US large language model (LLM) on Arena, a platform that ranks models through blind head-to-head comparisons, is Google’s Gemini 4 Carbon, with an estimated score of 1528, closely followed by Google’s Gemini 4 Argon at 1520 and Anthropic’s Claude Opus 5.5 at 1509. China’s top model, Moonshot’s Kimi K3, released in mid-July 2026, scores an estimated 1488, placing it 19th. The last top US model at this level was Google’s Gemini 3 Pro, which led from its release in mid-November 2025 until late January 2026. This gap is consistent with claims that the United States has a six-to-nine-month lead in AI capabilities.
The global spread of LLMs is arguably the most important front in the AI race. The models that consumers, businesses, and governments rely on will shape the information they see, how their economies run, and, potentially, the future of war. Which country leads depends on how users access the models.
There are six main ways to do so:
Reliable data exist for four of these: websites, apps, self-hosted open models, and aggregator platforms.
China holds a comfortable lead in self-hosted open models and aggregator platforms. In the third quarter of 2026, Chinese models were downloaded nearly 1.8 billion times on Hugging Face, compared with 500 million for US models, giving China 78 per cent of the two countries’ combined downloads. Such models particularly appeal to governments and businesses, which can run them on their own servers to keep data local and adapt them to their needs. On aggregator platform OpenRouter, Chinese models accounted for 55 per cent of all text requests in the week commencing 26 October 2026 — compared to the US share of 38 per cent.
However, most users access LLMs through websites, apps, and embedded AI. The United States has a commanding lead in websites and apps. US models drew more than 90 per cent of combined global traffic to US and Chinese AI websites. US models also accounted for more than 80 per cent of combined use of US and Chinese AI apps, based on Google Play downloads and, for China, where Google Play is unavailable, QuestMobile data.
Overall, most people who use AI rely on US models, giving the United States unmatched reach and greater influence over the technology’s global direction.
With no other country posing a challenge to their dominance in AI, the United States and China have an outsized influence over how the technology is governed globally.
Rival blocs are already taking shape. Washington launched Pax Silica in December 2025 to secure AI supply chains among trusted partners. Beijing answered in July 2026 with the World Artificial Intelligence Cooperation Organization (WAICO), which centres on AI governance standards and digital development. Twenty-nine countries signed its founding agreement, and eight more have since joined. Asian membership is divided: eight countries belong to WAICO, while six, including Australia, belong to Pax Silica. No country is a signatory to both initiatives.
Yet neither the United States nor China is widely trusted to regulate AI. Nor is the European Union, which Pew Research Center surveys found was the most trusted of the three in 2025 but now trails both. Attitudes across Asia vary, particularly in Southeast Asia. A majority of Indonesian respondents trust both the United States and China, while majorities in India and the Philippines favour the United States. In Malaysia, Singapore, and Thailand, majorities trust neither, whereas in Bangladesh, Pakistan, and Sri Lanka, China is more trusted. With the region split, Australia, as a Pax Silica member, is well placed to act as a bridge between the US-led bloc and countries seeking to balance AI safety with investment in their own capabilities.
Australia is already active in AI safety and standards, having published guidance on AI adoption and AI ethics principles. In 2025, it announced an AI Safety Institute to assess the capabilities and risks of emerging AI and publish its findings. Crucially, Australia’s economic and institutional ties to Asia are deeper than the European Union’s. Building on its governance and research leadership as well as its strong regional ties, Australia could help Asian countries develop their own AI models, policies, and regulatory guidelines. This would give them a shared foundation for AI governance without depending on either superpower.
This Data Snapshot builds on the Lowy Institute Asia Power Index. Rather than focusing on all 27 countries and territories, it compares only the United States and China, as no other country can challenge them in AI development. It comprises four measures — Resources, Technology, AI capabilities, and AI adoption — made up of 14 indicators. Two of those indicators, critical minerals production and reserves, are scored from a further 66 sub-indicators, covering the minerals for which data are available. Measures, indicators, and sub-indicators are all weighted equally at each level.
Four of the 14 indicators are drawn from the Asia Power Index, most from its technology sub-measure. To give a more comprehensive account of US–China AI rivalry, the Snapshot adds indicators covering areas the Index does not.
To show which country leads, each country’s value was divided by the two countries’ combined total. Each indicator is thus scored from 0 to 1, with 0.5 indicating parity.
The authors would like to thank Clare Caldwell, Ian Bruce, Mihai Sora, and Jack Xu for their contributions to this Snapshot.
About the authors
Jack Sato
Jack Sato is a Data Analyst for the Asia Power Index at the Lowy Institute.
Hunter Marston
Dr Hunter Marston is the Director of the Southeast Asia Program at the Lowy Institute and Project Lead for the Asia Power Index.
