In this paper, we focus on the impact of early grandparents' care on child cognitive outcomes, in the short and medium term, using data from the Millennium Cohort Study (UK). Compared with children looked after in a formal care centre, children cared by grandparents (as well as parents) are better in naming objects, but worse in tests concerning basic concepts development, problem-solving, mathematical concepts and constructing ability. In order to assess a causal link between early care and child outcomes, we employ panel methods and instrumental variables techniques that confirm that grandparental care matters more for naming ability while formal care is more important for problem-solving ability and basic concepts development. These results hide strong heterogeneities: on the one hand, the positive association between grandparents' care and child outcomes is stronger for children in more advantaged households; on the other hand, the negative association is significant only for children in more disadvantaged households.