The $3 trillion spending spree on AI datacenters is triggering a secondary, and equally massive, wave of investment: a $720 billion bill just to upgrade the world’s power grids to handle the new load.
The demand for AI is fueling a historic expansion of datacenters. Global capacity, now at 59GW, is set to double by 2030, according to Goldman Sachs. This year, work is set to start on 10GW of new capacity—a power draw equal to roughly a third of the UK’s entire national demand. Datacenters, typically operating at 60% capacity, are already power-hungry, and AI workloads are even more intensive.
This “incredible” $3tn spend on the datacenters themselves is being driven by “hyperscalers” like Google and Microsoft, who are spending $750bn in the next two years. Mega-projects like the $500bn “Stargate” are also planned.
But this digital infrastructure is useless without electricity. The Goldman Sachs report highlights the $720bn cost to upgrade grids, a cost that underscores the massive physical constraints of the AI revolution.
While debates rage about whether the $3tn spend is a “bubble,” the $720bn grid cost is a hard requirement. It represents the hidden-in-plain-sight cost of building an AI-powered future and raises significant questions about whether our energy infrastructure can keep pace with our technological ambitions.
30
previous post