Numerous urban indicators scale with population in a power law across cities, but whether the cross-sectional scaling law is applicable to the temporal growth of individual cities is unclear. Here we first find two paradoxical scaling relationships that urban built-up area sub-linearly scales with population across cities, but super-linearly scales with population over time in most individual cities because urban land expands faster than population grows. Different cities have diverse temporal scaling exponents and one city even has opposite temporal scaling regimes during two periods, strongly supporting the absence of single temporal scaling and further illustrating the failure of cross-sectional urban scaling in predicting temporal growth of cities. We propose a conceptual model that can clarify the essential difference and also connections between the cross-sectional scaling law and temporal trajectories of cities. Our model shows that cities have an extra growth of built-up area over time besides the supposed growth predicted by the cross-sectional scaling law. Disparities of extra growth among different-sized cities change the cross-sectional scaling exponent. Further analyses of GDP and other indicators confirm the contradiction between cross-sectional and temporal scaling relationships and the validity of the conceptual model. Our findings may open a new avenue towards the science of cities.