Who Will Win the AI Race: the US or China?

The latest figures show a clear split in the global AI race. The United States continues to lead on funding and hardware, while China has accelerated in research output and deployment. Stanford reports that US private AI investment reached $109.1 billion in 2024, almost twelve times China’s $9.3 billion. US firms also operate around 74 per cent of the world’s high-end AI supercomputing capacity, compared with roughly 14 per cent in China. At the same time, China is producing AI research at scale, publishing around 23,700 papers in 2024 and filing more than 35,000 AI-related patents.

The sections below examine how these trends play out across investment, talent, real-world deployment and more:

1. Compute: The Core Capability Engine
2. Semiconductors and Manufacturing: The Strategic Supply Chain
3. Investment, Innovation Ecosystems and Market Leadership
4. Research Output and the Talent Landscape
5. AI Model Performance: Narrowing the Gap
6. Industrial Deployment and National Strategy
7. Geopolitical Effects of the AI Race
8. Who Is Winning the AI Race?

Compute: The Core Capability Engine

Compute power, the processing capacity required to train and run advanced AI models, remains one of the clearest indicators of global leadership. As of 2025, the United States holds a decisive advantage, controlling roughly half of global AI compute through hyperscale cloud platforms and research supercomputers. This concentration allows US-based organisations to train larger models, iterate faster and set frontier benchmarks. It underpins the dominance of foundational platforms such as OpenAI and Google, whose access to compute helps define global performance standards.

Co-Founder and CEO of Google DeepMind, Demis Hassabis has repeatedly argued that scale remains fundamental to progress in advanced AI, noting that pushing current systems to their computational limits is essential for reaching more general intelligence. His perspective mirrors the US approach, where access to vast compute resources continues to shape global performance benchmarks (Source: WIRED).

Read more about Demis Hassabis and DeepMind here: Speaker Spotlight: Demis Hassabis – Co-Founder & CEO of Google DeepMind

(Source: The Federal Reserve)

In practical terms, the United States hosts far more physical infrastructure. By 2024, it had approximately 4,049 data centres, compared with around 379 in China. That year alone, the US added an estimated 5.8 GW of new data centre capacity, far outpacing Europe and the UK. On a per-capita basis, the gap is sharper still, with roughly 100 servers per 1,000 people in the US, versus fewer than 10 per 1,000 in China.

China’s compute ecosystem is smaller but increasingly adaptive. Export controls on advanced chips have pushed Chinese developers to prioritise efficiency, algorithmic optimisation and alternative architectures. A clear example is DeepSeek, founded in 2023 by Liang Wenfeng, also a co-founder of the quantitative hedge fund High-Flyer. DeepSeek has attracted attention for delivering competitive performance under hardware constraints and for releasing open-source models, demonstrating how strategic design choices can partially offset limited resources.

Read more about Liang Wenfeng and DeepSeek here: Speaker Spotlight: Liang Wenfeng – Founding CEO of DeepSeek

Kai-Fu Lee has described US chip restrictions as a catalyst for Chinese innovation, arguing that hardware limits have forced teams like DeepSeek to focus on algorithmic efficiency rather than brute-force scale. His analysis helps explain how China is narrowing performance gaps despite constrained compute access (Source: Bloomberg).

China continues to build data centres rapidly, often with local government backing, although many remain underutilised. The compute gap therefore reflects both scale and strategy. The US leverages raw capacity to drive frontier breakthroughs, while China concentrates on efficiency and targeted innovation.

Semiconductors and Manufacturing: The Strategic Supply Chain

If compute is the engine of AI, semiconductors are its fuel. This places chip leadership at the centre of strategic competition between the United States and China. American firms continue to dominate advanced AI chip design, with Nvidia processors underpinning most large-scale language models and data centre AI systems worldwide.

In late 2025, US policy signalled a partial shift. Washington allowed Nvidia to sell certain advanced chips, including the H200 series, to approved Chinese customers under strict conditions. The move softened earlier restrictions while adding a new layer of geopolitical complexity (Source: BBC). China, meanwhile, is investing heavily in domestic semiconductor capability. State-backed firms such as Huawei and Cambricon are central to initiatives like the Xinchuang strategy, which promotes homegrown technology across critical sectors (Source: Bloomberg).

Robert Hannigan, former head of GCHQ, often frames advanced chips as a national security asset rather than a purely commercial one. His work on cyber and geopolitical risk highlights why semiconductor supply chains have become a focal point of US-China strategic competition (Source: Robert Hannigan Website).

Semiconductors sit firmly within China’s long-term economic planning, with domestic AI accelerators now deployed at scale across public-sector data centres to improve resilience and self-sufficiency. Despite this momentum, structural challenges remain. Chinese-designed AI chips still lag behind leading US alternatives, while fabrication capacity is constrained by limited access to advanced manufacturing tools (Source: Council on Foreign Relations).

Advanced chipmaking remains China’s weakest point. While the 2022 US CHIPS and Science Act committed $52 billion to domestic fabrication, and China followed with a $47.5 billion national chip fund in 2024, the technology gap persists. At the cutting edge, Taiwan Semiconductor Manufacturing Company produces around 92 per cent of the world’s most advanced chips at five nanometres or below, while China’s SMIC reached seven nanometres in 2022 (Source: GIS Reports).

In practical terms, performance differences remain stark. A recent CFR study shows Nvidia’s H200 delivers around five times the performance of Huawei’s most advanced current AI chip, with the gap expected to widen. At the same time, global supply chains remain deeply interwoven. US chip designers rely on manufacturing partners in Taiwan and South Korea, while Chinese components continue to be embedded across global infrastructure (Source: McKinsey & Company).

Yoshua Bengio has noted that hardware constraints increasingly shape the direction of AI research itself, influencing which models are viable at scale. His work reinforces why chip access is not just an industrial issue, but one that directly affects the trajectory of AI capabilit (Source: MIT Technology Review).

Investment, Innovation Ecosystems and Market Leadership

(Source: Stanford)

Capital and innovation ecosystems play a decisive role in shaping who builds AI at scale and who commercialises it globally. The United States continues to attract the majority of private AI investment, with US startups receiving far more venture funding than their Chinese counterparts across 2024 and 2025 (Source: Stanford). That capital is largely directed towards foundational models, enterprise software and generative applications, supported by strong demand from cloud providers and multinational businesses. This creates an environment where experimentation, scaling and monetisation progress quickly (Source: S&P Global).

Sam Altman’s ability to raise unprecedented levels of private capital for OpenAI illustrates how US funding ecosystems accelerate experimentation and global deployment. His work shows how venture capital, cloud partnerships and research labs combine to convert frontier research into widely used products (Source: Bloomberg).

China’s investment landscape follows a different path. Funding is more state directed, with development banks and municipal programmes channelling capital into AI infrastructure, semiconductor manufacturing and industrial automation (Source: CNN). The funding gap remains significant. The Federal Reserve confirms that the US accounts for the majority of global AI venture capital, including more than 75 per cent of funding for generative AI startups. Around $100 billion was invested in AI companies worldwide in 2024, up from $55.6 billion in 2023, with US firms securing the largest deals.

Azeem Azhar often describes this contrast as one between market-led innovation and state-coordinated execution. His analysis helps explain why US firms dominate global platforms, while China excels at large-scale domestic implementation (Source: Harvard Business Review).

China’s smaller private VC market is partially offset by state-backed capital. Local governments have invested an estimated $184 billion into nearly 10,000 AI startups over the past decade, alongside a newly announced $138 billion long-term fund for AI and quantum technologies (Source: Stanford). Even so, 73 per cent of large language models originate in the US, compared with around 15 per cent in China, underlining the continued commercial dominance of the US ecosystem (Source: Eurasia Review).

Research Output and the Talent Landscape

AI leadership ultimately depends on people. Over the past two decades, China has moved from a marginal contributor to a dominant force in global AI research. By 2025, this shift reflects sustained investment in education, expanded university capacity and clearly defined national priorities. The United States remains a powerful magnet for elite talent, with universities and industry labs continuing to attract global researchers and produce a high share of influential work.

Dr Fei-Fei Li frequently highlights the global nature of AI talent, noting that many of the field’s most influential researchers have been trained in US institutions before contributing internationally. Her perspective underscores why talent mobility remains a strategic advantage for the US (Source: WIRED).

The gap between the two ecosystems is more nuanced than headline figures suggest. China leads in research volume and participation, while the US retains an edge in high-impact outputs and in translating academic breakthroughs into commercial systems. Strong links between universities, venture capital and industry have helped sustain this pipeline. China’s scale is striking, with more than 38,000 generative AI inventions over the past decade, around six times the US total.

Talent presents a mixed picture. Around half of AI PhD students in US universities are international, and US institutions still host many of the most cited researchers (Source: The White House). China now produces more AI bachelor’s and PhD graduates than the US, and output is growing faster. At the same time, a reverse brain drain is emerging, as increasing numbers of China-born scientists return home, drawn by pay, prestige and stability.

AI Model Performance: Narrowing the Gap

Model capability has become a central arena of competition. US-based models have historically outperformed Chinese counterparts on standard benchmarks, yet recent results show the gap narrowing. Between early 2024 and early 2025, comparisons point to clear convergence, particularly where Chinese models demonstrate strong efficiency under resource constraints.

Yann LeCun has long argued that open research accelerates global convergence, making sustained performance gaps difficult to maintain. His view helps explain why Chinese models are increasingly matching US benchmarks within short timeframes (Source: Forbes).

The United States continues to lead on raw capability, but China’s trajectory now signals meaningful competitiveness, especially within its domestic market. This reflects growing technical maturity and a sharper focus on aligning model development with deployment needs rather than scale alone.

(Source: Stanford)

Stanford’s AI Index recorded 40 notable new models from US-based teams in 2024, compared with 15 from China. By late 2024, Chinese models had effectively matched US systems on key benchmarks. On demanding tests such as MMLU and HumanEval, score gaps fell from double digits in 2023 to near parity. DeepSeek v3 illustrates this shift, matching many US benchmarks at a fraction of the training cost (Source: Stanford).

Industrial Deployment and National Strategy

China’s national AI strategy prioritises scale and deep integration with the real economy. Manufacturing, logistics and infrastructure increasingly depend on AI-driven automation, predictive optimisation and real-time analytics, supported by policy incentives and coordinated deployment.

This approach embeds AI within traditional industries and public services, underpinning long-term productivity growth. In contrast, US adoption is led by digital services, enterprise software and cloud platforms serving global markets. These approaches reflect complementary strengths: China in industrial integration and delivery, the US in software ecosystems and consumer-facing AI products.

Manufacturing highlights the contrast most clearly. Global industrial robot installations reached around 4.28 million units in 2023. China installed 276,288 robots that year, accounting for 51 per cent of the global total. By comparison, the US installed approximately 37,600 units in 2023. China’s robot density has increased fourfold since 2017, underlining the speed and scale of factory automation. Martin Ford has warned that AI-driven automation will reshape global manufacturing incentives, reducing reliance on low-cost labour. China’s rapid industrial deployment aligns closely with his forecasts on automation-led productivity (Source: Fortune).

(Source: Stanford)

AI adoption across services is also accelerating. In 2024, 78 per cent of organisations worldwide reported using AI, up from 55 per cent in 2023. In healthcare, the US approved 223 AI-enabled medical devices in 2023, compared with just six in 2015. Autonomous mobility is scaling too. Waymo delivers around 150,000 robotaxi rides per week, while Baidu’s Apollo Go operates across multiple Chinese cities (Source: Bloomberg).

Geopolitical Effects of the AI Race

AI competition cannot be separated from geopolitics. In the United States, AI leadership is framed as an issue of economic sovereignty and national security, shaping export controls, trade negotiations, alliances around chip manufacturing and data governance policy. China has responded by prioritising technological self-reliance, promoting domestic hardware for government data centres and investing in national champions to reduce external dependence.

Despite this rivalry, interdependence persists. US firms continue to explore limited access to China’s market under strict conditions, while global supply chains remain deeply connected. Competition and cooperation coexist, reinforcing that the AI race is a complex strategic relationship rather than a simple duel.

Who Is Winning the AI Race?

There is no single answer. Leadership in AI spans compute, capital, talent, research, commercialisation, deployment and geopolitics. The United States leads in compute capacity, venture capital, global platforms and high-impact research, while China leads in deployment scale, domestic coordination and industrial integration.

Andrew Ng has cautioned against framing AI as a zero-sum race, arguing that its benefits will diffuse globally regardless of national boundaries. His view supports the idea that AI leadership will remain multipolar rather than absolute (Source: Harvard Business Review).

The US–China AI race is a long-term strategic contest rather than a short sprint. The United States benefits from structural advantages in capital, compute and global reach, while China leverages scale, coordination and rapid deployment. Global AI leadership is likely to be multipolar, shaped by diverse ecosystems that balance innovation, deployment and responsibility.