Should we tax artificial intelligence?

I study the optimal taxation of artificial intelligence (AI). Consistent with recent empirical evidence, I model AI as a technology that substitutes for high-skill labor. The main theoretical result is that AI should be subsidized in the second-best optimum regardless of the government’s preference for redistribution. This result contrasts sharply with other automation technologies, for which the optimal tax is typically positive or ambiguous. This is because AI adoption compresses the wage distribution and therefore relaxes incentive constraints that limit redistribution through income taxation. In the calibrated model, the optimal AI subsidy rate is 0.61%. Its welfare impact is modest, with most welfare gains arising from adjustments to labor income taxes.