Compute-to-GDP Fallacy skews AI pacing debate

The authors frame the current AI pacing debate as a distraction from the real economics of enterprise adoption. Drawing on more than 100 conversations with CEOs, policy leaders, and AI scientists for their coming book, they argue that the structural realities of enterprise architecture—fragmented data silos, legacy ERPs, strict compliance regimes, and basic data hygiene—make true economic absorption inherently slow regardless of what frontier labs decide. The McKinsey data they cite reinforces this: only 6 percent of companies reported a "significant" impact and modest earnings attribution from AI, and electricity took 75 years to lift productivity economy-wide.

The piece also notes a parallel shift in silicon economics. Older-generation chips are finding new life for practical inference tasks, with Growth Protocol founder Miro Dimitrov reporting at the Yale CEO Caucus that deploying neuro-symbolic architectures slashed inference costs by roughly 80-fold in live client deployments. This dynamic suggests that the frontier labs' single-minded race toward superintelligence may be less commercially decisive than they assume, since most enterprise demand can be met by models one or two generations old.

The stakes the authors identify hinge on whether the labs recognize that two races are underway: one toward superintelligence and another for share of wallet. The outcome may depend on whether frontier labs can earn the trust of enterprises, regulators, and citizens—since recapturing a customer after the decision is made is extraordinarily difficult, as China's victory in the global telecommunications race illustrates. A pacing interval, in their view, is less an economic ceasefire than an active defensive hardening window.