Abstract：I will discuss the advantages of observing the Universe through 21cm intensity mapping. I will point out that in order to extract the maximum cosmological and astrophysical information from 21cm surveys such as Tianlai, FAST, and SKA, accurate theoretical predictions for the distribution of the cosmic signal are needed. I will present the limitations of using linear theory to achieve that, and will show how we can use state-of-the-art magneto-hydrodynamic simulations to model the distribution of HI from linear to fully non-linear scales. I will show the most sophisticated, and computationally expensive, 21cm maps created to-date, and present the ingredients needed to reproduce them without running expensive hydrodynamic simulations. I will show how 21cm observations can be used to learn about the efficiency of astrophysical processes and to constrain the nature of dark matter and the sum of the neutrino masses. I will present results from simulations that contain the cosmological signal, system noise and foregrounds, to show how well we can determine the position of the BAO peak with SKA.