Positive effects of habitat patch size on biodiversity are often extrapolated to infer negative effects of habitat fragmentation on biodiversity at a landscape scale. However, such cross-scale extrapolations often fail, undermining the effectiveness of policies based on them. Forty-two percent of the manuscripts that cited a recent, landmark patch-scale analysis (Chase et al. 2020, Nature 584, 238–243) used it to infer negative fragmentation effects, despite the fact that Chase et al. cautioned against such extrapolation. We use the same dataset used by Chase et al. to test whether it is valid to extrapolate their observed patch-scale patterns to a landscape scale. We compare species diversity across many small vs. few large patches totaling the same area and sampling effort, while accounting for the patch-scale "ecosystem decay" observed in Chase et al. We find that landscape-scale patterns are opposite to their equivalent patch-scale patterns. For an equal total habitat area, species richness and evenness decrease with mean size of the patches comprising that habitat area, whether considering all species or only species of conservation concern. Thus, landscape-scale processes (e.g., turnover) enhance biodiversity across sets of many small patches despite ecosystem decay within each patch, and this is not due to incursion of generalist species into small patches. Our results confirm that patch-scale biodiversity patterns can rarely be extrapolated to a landscape scale. Most importantly, they highlight that conservation should recognize the value of small habitat patches as an important asset for preserving biodiversity worldwide.