I explain it this way; think of it as a cellular network in the sky, except in a terrestrial cellular network, the cell sites are stationary and connected via fiber and you are moving in your car or walking (or you can). With LEOS, each satellite is like a cell site, like a 5G one with phased array antennas, the satellite is moving and you are stationary and the satellited are connected to each other via free space optics but the backhaul is obviously RF, but the hand offs between satellites is over FSO. In fact, R17 of 3GPP standards now has provisions in it for 5G direct from the LEOS to the user. None deploy yet. but in a couple of years, that could be interesting.

Each satellite paints a spot on the ground using the beamforming capabilities of the phased array antenna. The capacity of that spot is of course limited by Shannon limit, where C ~ B * S C = capacity (think Mbps) B = spectrum (think Mhz) S = SINR (Singal to noise ratio)

If you have a smaller spot on the ground, then the spectrum is shared amoung fewer users. If you have more powerful transmitters and or better antennas, then your C will be better.

Right now based on the specifications I have seen publicly, Amazon's Kuiper (their LEOS) will have higher capacity than starlink because it will have half the "spot" diameter. That is due to better antennas mostly, more array elements mean narrower beam width.