Focus
Artificial intelligence (AI) has the potential to substantially boost economic growth. When transformative, AI could automate an expanding set of tasks, improve machines’ capability and unlock the endogenous growth engine. Some recent estimates suggest growth could exceed 10% within a few years under such transformative scenarios. At the same time, large-scale automation could absorb resources, diverting more output to investment at the expense of consumption. Could AI sustain higher growth even as households buy a smaller share of total output?
Contribution
We examine demand bottlenecks as a possible constraint to otherwise transformative AI. We employ a task-based endogenous growth model where innovation is driven by rents on consumer products. As automation advances, the consumption share and the associated rents decline, weakening incentives to innovate further with AI. The framework allows for several scenarios depending on how fast AI technologies arrive and how long rents last, among other factors. Evidence on these parameters provides discipline in assessing the scenarios. The model also has implications for the natural rate of interest and the broader AI debate.
Findings
We identify several AI scenarios. Under “transformative AI”, growth takes off, reaching 5.5% after two decades and continuing to rise. Under “demand bottleneck”, growth accelerates briefly before falling below the pre-AI rate of 2% as innovation permanently slows. Other scenarios similarly point to, at best, moderate gains in growth, far below the transformative AI outcome. The calibrated model points to the demand bottleneck scenario as empirically highly plausible. The range of possible outcomes underscores the high uncertainty about the natural rate of interest.
Abstract
Transformative AI promises to accelerate growth through productivity enhancement and automation. Yet automation on such a scale would raise the share of investment in output, potentially starving AI innovation of returns from a shrinking consumer market. We propose an endogenous growth framework to study this interaction, featuring research and entry decisions that propel AI advances but whose costs must be paid for by consumer demand. We identify several scenarios, spanning transformative AI, a demand bottleneck outcome where thinning consumer demand stalls AI progress, and intermediate cases in between. When anchored to evidence, the calibrated model points to the demand bottleneck as a highly plausible scenario, and projects more moderate growth in the coming years relative to recent studies such as Korinek et al. (2026). We draw policy implications, including for the natural rate of interest.
