Nvidia's Jensen Huang just dismantled every AI skeptic with one simple equation to justify $5 trillion in AI infra spending by 2030. I have visualized it below.
He asserts human cognitive work is roughly $50 trillion of global GDP. If AI makes that labour more productive by a material amount, he posits an added output of about $10 trillion by 2030. With roughly 50% gross margin on those AI services, about $5 trillion becomes the underlying infrastructure costs through data centres, chips, networks, and energy.
This year, $400 billion will be spent on AI capex by the big 5 hyperscalers (Meta, Amazon, Alphabet, Microsoft and Oracle).
Jensen says productivity can jump 2-3x by investing $10,000 in AI per $100,000 employee and that he's already seeing this in action across Nvidia's engineers and chip designers. So underbuilding is the bigger risk.
However, these numbers are based on future productivity promises, not current evidence.
While the 2030 timeline makes Jensen's argument more plausible, it's still extremely aggressive. It requires ~50% annual productivity improvements every year when current data shows we're barely achieving that in pilot programs.
Jensen assumes augmentation (workers + AI = 2-3X output), but evidence seems to suggest substitution where AI is replacing, not enhancing, many knowledge worker tasks. GDP impact is redistributive, not additive.
(The visualization was made with Claude Code!)