The antennas are practically free; the cost comes from phase shifters, amplifiers, etc, which don't see much miniaturization benefit at high frequencies. One day $1 per element is feasible, but definitely not today.

I'm betting the front-end chip is all CMOS, as opposed to SiGe, so it will be quite cheap (under $10 in volume). 16 TX elements at a total of 20 dBm transmit power is 8 dBm/element, which is feasible in CMOS. State of the art in CMOS is 9 dBm at 80 GHz.

The antenna will be cheap too as there are number of spread spread glass laminates out now; I have used Rogers RO1200 for mmWave work. The only thing that may make it expensive is the air-gap construction, which they have probably done to increase the bandwidth. I would assume two laminates are needed.