Chip → Grid: Why engineering, not capital will decide who wins the AI infrastructure boom
September 2026 | Market thesisCapital is abundant, but engineering capacity is not. Six bottlenecks stand between the chip and the grid, and ER&D firms that move early on power electronics, talent and interconnect validation can capture the opportunity.
Executive Summary
Capital for the AI infrastructure build-out is abundant, global data-centre capex is tracking toward $7T by 2030. What's scarce, and what this report is about, is the physical and engineering capacity to convert that capital into energized, cooled, connected compute. Global data-centre power demand is set to rise from 132 GW in 2026 to roughly 290 GW by 2030, while grid interconnection takes up to five years and heavy electrical equipment lead times run 2.5–4 years. Six bottlenecks stand between capital and energized compute, power delivery, the engineering talent gap, the gallium chokepoint, copper economics, standards churn, and financing discipline and only one of them, power delivery, is getting real market attention.
Within the stack, power electronics/800VDC carries the most urgency, moving from specification to commercial deployment this year; interconnect validation carries the most durable, recurring economics, the closest thing to annuity revenue in this stack.


