AI spending boom faces trillion‑dollar productivity question

The numbers in this piece put hard bounds on the current AGI gold rush. Trillions in projected capex for data centers, power and networking, plus Anthropic’s own plan to spend more than 100 times its 2025 revenue on future models, only make sense if AI delivers multi‑percentage‑point annual productivity gains across major economies. Economists quoted by Reuters point out that such broad gains are not yet visible, and that the historical lag between general purpose technologies and measured productivity can be decades.

In the near term, this disconnect creates a strange dynamic: AGI labs are capital‑starved on a multi‑trillion dollar scale, yet equity investors are already stretched on valuations that assume near‑term transformation. If the killer applications that can absorb hyperscaler‑scale capacity arrive quickly, the resulting cash flows will let labs fund more aggressive frontier bets and hardware buildouts. If they do not, over‑levered infrastructure owners and single‑lab‑heavy portfolios could face a nasty correction.

For the race to AGI, that means timelines are partly financial. A sustained boom in AI infra spending without an economic crash would effectively subsidize multiple shots at very large models and agent platforms. A sharp pullback, by contrast, could slow the cadence of extreme‑scale training runs and push more effort toward efficiency, smaller models and targeted verticals rather than pure frontier capability.