The Compute Supercycle: Infrastructure, Geopolitics and the Structural Transformation of Global Growth

Abstract

From how AI hardware and energy systems intersect, to the global economy's shift toward more resource-intensive servers and low-scale computing, this paper argues that this “compute supercycle” is not just a trend in tech spending, but a reinvention of the global economy. We create a single model, compute-economics, that brings together three blocks: AI capital expenditure, semiconductors supply chain, data-center energy usage and project, region and industry specific AI productivity gains, leveraging primary data from IMF, World Bank, IEA, WSTS, IDC, FRED, Goldman Sachs, Morgan Stanley and World Economic Forum. On the practical front, we show that in 2026 annual AI CAPEX will be 1.4 trillion, with total capacity growth in data centers expected to be 847 GW and the revenue from semiconductors will be 912 billion in 2026.Empirically, for instance, you can see that in 2026 the annual investment in AI will be 1.4 trillion, the power consumption of data centers will be 847 GW, and the AI's semiconductors revenue will be 912 billion. Extreme concentration of fabrication in Taiwan, energy constraints degreeed as more and more data centers drive electricity consumption, and the more import of advanced chips are controlled – three risk factors get studied and we try to gauge their likely effect on both bull, base and bear scenarios for compute demand, semiconductors and energy infrastructure buildup. Methodologically, we introduce the Compute Economics 2026 Dashboard, a multi tab visualization and analytics platform that is real time and becomes the platform for operationalising this framework for investors, policy makers and corporate strategists; bringing different types of heterogeneous macroeconomic, market and technical indicators together in one decision interface.