Large arrays of neutral atoms are a leading modality in the race towards useful quantum computation. Systems with more than 1000 qubits in 2D arrays and gate fidelities approaching three nines are being developed by many academic and industrial groups. I will present recent progress at University of Wisconsin-Madison and Infleqtion with physical and logical qubit encodings. Defining an optimal architecture that maximally reduces both space and time overheads is an active research challenge. I will describe Rydberg based, motion based, and optics based approaches to connectivity at short, medium, and long length scales. A hybrid solution that combines local optical control with short distance motion appears optimal for the surface code.