The late-time integrated Sachs–Wolfe (ISW) signal in the CMB temperature anisotropies is an important probe of dark energy when it can be detected by cross-correlation with large-scale structure surveys. Because of their huge sky area, surveys in the radio are well-suited to ISW detection. We show that 21cm intensity mapping and radio continuum surveys with the SKA in Phase 1 promise an ∼5σ detection if we use tomography, with a similar forecast for the precursor EMU survey. In SKA Phase 2, the 21 cm galaxy redshift survey and the continuum survey could deliver an ∼6σ detection. Our analysis of the radio surveys aims for theoretical accuracy on large scales. First, we include all the effects on the radio surveys from observing on the past light-cone: redshift-space distortions and lensing magnification can have a significant impact on the ISW signal-to-noise ratio (SNR), while Doppler and other relativistic distortions are not significant. Secondly, we use the full information in the observable galaxy angular power spectra Cℓ(z, z′), by avoiding the Limber approximation and by including all cross-correlations between redshift bins in the covariance. Without these cross-bin correlations, the ISW SNR is biased.