We use census data at the U.S. county-level to provide causal evidence of the long-term health benefits of COVID-19 vaccines for elderly vis-à-vis younger adults. We use triple-difference estimates that control for all variation at the level of each (county, time) pair to exploit variation across age groups within each such pair. We also undertake difference-in-difference tests exploiting quasi-random variation in vaccination across counties over time separately for elderly and younger adults. A one standard deviation increase in vaccination decreases deaths per million one year later by 9 for the elderly and 1.9 for the younger adults. Similarly, for ICU admissions and hospitalizations per million, the long-term effects are disproportionately larger for elderly than for younger adults. In contrast, for cases per million, the long-term effect on elderly is about half that on younger adults. Such long-term effects of two doses cannot be attributed to the booster dose now.