Abstract Radiative forcing (F) $(F)$ is the fundamental driver of ocean heat uptake (N) $(N)$; however, its contribution to intermodel spread in N $N$ has not been quantified. We address this gap by leveraging a new set of CMIP6 experiments that diagnoses time‐varying F $F$ for the first time in 11 models. Among them, we find that intermodel spread in N $N$ is strongly affected by F $F$ in historical simulations, but by model sensitivity ν=N/F $nu =N/F$ in future projections. As a result, the correlation between historical N $N$ and future N $N$ is highly uncertain (0.1–0.7), whereas the correlation between historical ν $nu $ and future ν $nu $ is much more robust (0.8–0.9). For the 11 models, we show that imposing observed ν $nu $ to constrain future N $N$ can narrow its likely range by at least 10%. Finally, we emphasize that formulating N $N$ as N=νF $N=nu F$ opens up new avenues for future ocean heat uptake studies.