Introduction
Five Cents looks at Who Leads the Global AI Race by separating the headline contest from the machinery behind it. We’ll compare frontier models, chips, cloud capacity, talent, products and regulation. That means looking at the United States and China, as well as Taiwan, South Korea and Europe.
The best place to begin is with a simple correction
The best place to begin is with a simple correction: there is no single AI leaderboard. A company can build a leading model while relying on another company’s chips, another country’s factories and a cloud provider’s data centres. A country can lead in patents, industrial deployment or regulation without producing the most popular chatbot.
So the real question is not who is first overall. It is who controls the layers that turn research into useful, scalable systems. Start with models. US companies remain especially strong in frontier systems, private investment, startup creation and global product distribution. OpenAI, Google, Anthropic, Meta, Microsoft and Amazon combine advanced research with large user bases and cloud infrastructure.
China is much closer than a simple “first versus second” framing suggests. Alibaba, Baidu, Tencent, Huawei, DeepSeek and others have built capable models, large domestic platforms and extensive industrial deployment. The gap on some headline model tests has narrowed sharply. But benchmarks are snapshots, not proof of broad, real-world reliability.
Then comes the physical bottleneck: compute. NVIDIA has built the dominant accelerator platform. But AI hardware is not just about the fastest GPU. It is a system of processors, high-bandwidth memory, networking, software, cooling and electricity. Taiwan’s TSMC is central because it manufactures many of the most advanced chips. The Netherlands matters because ASML’s lithography equipment enables leading-edge chip production. South Korea is pivotal through memory suppliers such as SK hynix and Samsung.
These are not supporting roles
These are not supporting roles. If any of these links fail, model developers cannot simply code their way around the shortage. Cloud and energy are becoming just as strategic. Training a frontier model and serving millions of users require enormous data-centre capacity. US cloud providers have a major advantage in global scale, while Chinese firms operate vast domestic infrastructure and are building alternatives under export restrictions.
But more capacity depends on grid connections, power generation, cooling and permits. In the next stage of the race, access to reliable electricity may matter almost as much as access to algorithms. Products decide whether technical capability becomes influence. Search, office software, coding tools, social platforms, phones, factories and logistics networks generate users, revenue and feedback. That gives large platforms an advantage. They can test AI in real workflows and improve it based on what actually succeeds or fails.
China is particularly strong in integrating AI into e-commerce, logistics, manufacturing and public-facing services. Europe has fewer consumer-facing frontier-model champions. But it has important industrial players, research institutions, specialised engineering and a critical semiconductor-equipment position.
Regulation belongs in this comparison too. The European Union is shaping a comprehensive, risk-based framework, with obligations that affect how some systems are documented, tested and deployed. The United States uses a more fragmented mix of agencies, procurement, standards, export controls and state rules.
China combines industrial policy with data
China combines industrial policy with data, security and content controls. None of these approaches is simply more or less regulated. They pursue different balances between market access, safety, state control and national capability. The common mistake is to treat announcements as achievements. When a company unveils a model, chip or national strategy, ask five things: what was announced, what is available now, what has been independently measured, under what conditions, and what changes in practice.
A model demo may be impressive but expensive or unreliable. A chip’s peak performance may not translate into cheaper training. A national investment target may not produce working data centres, skilled teams or deployed products. So the leaders are different at different layers.
The United States has the strongest commercial full-stack ecosystem. China has scale, research volume and deep industrial deployment. Taiwan, the Netherlands and South Korea hold indispensable positions in the semiconductor supply chain. Europe influences rules, industrial AI and key technologies. The lasting advantage will go to ecosystems that can connect all those pieces, not just produce the flashiest demo.
To explore this further, you can generate Five Cents episodes on AI Chips: The Supply Chain Behind the Boom and How AI Regulation Changes Global Competition. And with that, you're up to speed in a few minutes.