The Economic Clock
Can rapidly evolving AI economics change the timetable of an infrastructure investment cycle?

What this investigation will test
The companion editorial, What If AI Is Changing Its Own Economic Clock?, raises a testable question: can the simultaneous evolution of computing efficiency, AI adoption, agent-generated workloads, and capital financing materially change the timing of an AI infrastructure correction?
Arthur Hayes argues that the growth rate of announced AI infrastructure spending could slow in mid-to-late 2027, becoming clear in 2028. He recognizes Jevons paradox and distinguishes growing AI demand from the possibility of overbuilt or financially impaired infrastructure. The specific timetable, rather than a simplified pro-versus-con debate, is the subject of the proposed investigation.
The competing explanations
Demand outpaces efficiency: Lower costs unlock enough valuable workloads to absorb added capacity.
Efficiency and construction outpace monetization: Rapid innovation, excess supply and financing burdens weaken returns even as AI adoption expands.
Both occur: Total usage grows while older accelerators, individual facilities or leveraged capital structures lose economic value.
The investigation will independently test each possibility and seek evidence that could disprove the working hypothesis.
Research questions and evidence standards
The planned evidence review will compare like-for-like AI task costs against workload intensity; separate human adoption from autonomous-agent activity; examine actual paid usage and infrastructure utilization; and track energy, construction, depreciation, refinancing and replacement economics. Competing hypotheses, contradictions and material unknowns will be retained, not smoothed into a predetermined outcome.
No probabilities, report findings or investment conclusions are presented as completed research on this page.
MAi Conclusion — Research Pending
The full MAi meta-analysis, reconciliation and validation have not been completed. Accordingly, no final MAi Conclusion has been determined. The completed report will be provided through this stable research page when available, following publisher review.
