"If Singapore doesn't want to depend on data centres in North Virginia, it must build its own, at no small cost." [Singapore]
"McKinsey reckons sovereign AI could account for $600bn of annual spending by 2030, sped by local regulation on data handling" [local regulation on data handling]
The article frames sovereign AI as rational risk mitigation (avoiding single-region dependency like Middle Eastern oil) but economically irrational at the system level—deglobalisation is 'expensive for individual countries, but a windfall for their suppliers.' This dynamic generalizes beyond AI: any critical technology where governments prioritize autonomy over efficiency will generate duplicate capex windfalls for suppliers. The $600bn McKinsey estimate and Nvidia's $30bn sovereign revenue anchor this as a material structural shift, not marginal.
Data localization regulations (GDPR-style, China's data residency rules, India's data localization proposals) create a structural constraint that makes centralized global compute architecturally impossible. This differs from voluntary sovereign AI adoption: regulatory mandates eliminate the option of cost-efficient consolidation. The $600bn figure is thus partly a regulatory tax on global efficiency, with suppliers capturing the margin on redundant infrastructure.