Collective[i]'s Roy: AI missing from GDP until firms build, not cut

Roy grounds his argument in a historical pattern. He recalls economist Robert Solow's 1987 line that the computer age was visible everywhere but in the productivity statistics, and notes it took another decade of investment before US productivity growth broke out. He applies the same lens to AI, pointing out that while enterprise spend on generative AI ran to $30 to $40 billion in 2025, the productivity statistics show almost nothing.

He also leans on the productivity J-curve framework from Erik Brynjolfsson, Daniel Rock, and Chad Syverson, which holds that firms first take productivity losses as they invest in unmeasured intangibles like process redesign and retraining, then post gains later. Roy argues that data center construction is one exception that does show up in GDP as private nonresidential fixed investment, with hyperscalers alone pouring more than $300 billion into data centers, chips, and grid capacity in 2025. But he says that spending reflects infrastructure build-out rather than the broader productivity gains from AI adoption.

The piece distinguishes between cost-cutting and building as phases of adoption. Roy contends that the real economic impact will come when companies stop optimizing existing processes and start using AI to create new products and markets, which he expects to happen by 2029 according to his forecast that the share of transformational deployments will rise from 5% to 50%.