The common practice for GDP nowcasting in a data-rich environment is to employ either sparse regression using LASSO-type regularization or a dense approach based on factor models or ridge regression, which differ in the way they extract information from high-dimensional datasets. This paper aims to investigate whether sparse plus dense mixed frequency regression methods can improve the nowcasts of the US GDP growth. We propose two novel MIDAS regressions and show that these novel sparse plus dense methods greatly improve the accuracy of nowcasts during the COVID pandemic compared to either only sparse or only dense approaches. Using monthly macro and weekly financial series, we further show that the improvement is particularly sharp when the dense component is restricted to be macro, while the sparse signal stems from both macro and financial series.