How much investments do AI firms really need to meet soaring demand for data center capacity by the end of the decade? USD 2 tn, Bain and Co. said in its annual Global Technology Report (pdf). Demand is rising at more than twice the pace of Moore’s Law — a common benchmark that assumes the number of transistors on an integrated circuit doubles about every two years. Meeting this level of demand could require 200 GW of new capacity worldwide by 2030 — half of which will need to be in the US — and as much as USD 500 bn in annual data center investment.

The revenue gap: Even if companies redirected all IT budgets to the cloud and recycled AI-driven savings into infrastructure, Bain sees an USD 800 bn annual shortfall in the revenues needed to fund that buildout.

Hyperscalers are already committing record sums, with Microsoft, Amazon, and Meta expected to push combined AI spending above USD 500 bn a year by early next decade, Bloomberg Intelligence estimates. Yet, without new revenue models or major efficiency gains, the industry risks stranded infrastructure on one side or unmet demand on the other, Bain cautions.

Still, there are some wildcards. Breakthroughs in algorithms, hardware, or quantum computing could ease the crunch, but supply-chain chokepoints — from GPUs to electricity grids — remain critical risks.