Terrestrial biosphere models (TBMs) disagree on how the terrestrial carbon sink responds to climate futures. On climate-change timescales, this response is primarily regulated by the competing responses of photosynthetic carbon uptake to increasing CO2 and respiratory carbon loss to increasing temperature, among others. Whereas emerging photosynthetic proxies bring about increasingly robust biome-scale constraints on photosynthetic responses, the response of respiration to climate remains highly uncertain due to the lack of observational constraints directly informing biome-scale responses. Here, we leverage in situ atmospheric CO2 observations from a network of tall towers over North America and estimates of ecosystem respiration from an ensemble of TBMs and data-driven models to derive the biome-scale temperature sensitivity of ecosystem respiration. We find that a large subset of models shows higher biome-scale temperature sensitivities of ecosystem respiration than atmospheric observations suggest. We further show that correcting the temperature sensitivity of ecosystem respiration improves these models’ ability to explain observed atmospheric CO2 variability. This inconsistency between bottom-up representations of respiratory processes and the observed overall biome-scale responses highlights an urgent need to evaluate parameterizations of respiratory processes directly on biome scales, leveraging top-down observational constraints. Given the lower biome-scale temperature sensitivity of ecosystem respiration than those represented in a large subset of TBMs, the terrestrial carbon sink may be more resilient to warming than expected from model ensemble behavior.