We estimate the effect of grant aid on City University of New York (CUNY) students' borrowing and attainment using a regression discontinuity/kink design based on the federal Pell Grant formula. Each dollar of grant aid reduces loans by $1.80 among borrowers. We only find crowd-out of this magnitude in colleges that, like CUNY, "offer" no loan aid and require students to opt into borrowing. We develop and empirically support a model that shows opt-in or other fixed borrowing costs can lead grants to crowd out large amounts of loan aid, lowering some students attainment by reducing their liquid resources.