For three years, the global AI contest has focused on building ever larger models, with the US having OpenAI, Anthropic and Google; China having DeepSeek, Alibaba and ByteDance; and Europe busying itself with rules. However, some institutions with advanced models have not seen expected productivity gains, while some economies that build no foundation models are turning AI into practical capability. Singapore’s position may be more favourable than it looks due to structural conditions. Advanced models are becoming steadily easier to obtain, and the barrier to acquiring the technology is falling quickly. What separates leaders from the rest is whether they can make AI actually work; many deployments fail because the organisation is not ready, with processes not redesigned, blurred responsibilities, data not flowing, and staff unsure how to work alongside the machine. The greatest challenge AI brings is often institutional and organisational innovation. This capability might be called a country’s “AI operating system”, a body of institutional infrastructure including digital infrastructure, systems for training talent, regulatory frameworks, industry standards, organisational governance and mechanisms of social trust.