It's a round number, and assumes about 10% loss. But also, they can't make or launch the sats for $1.25 million each yet, so it might not be a bad number.
Also, by my math, 5:1 oversubscribed is about breakeven, and it appears to be what Starlink does (reports of 10-50 mbps by users).
Just to note, 5:1 oversubscription wouldn't give you 10-50mbps unless everyone was using the connection at full rate at once. That's why (quick google) DSL services often oversubscribe by up to a 100:1.
The biggest factor driving disagreement seems to be satellite lifetime. You also didn't account for oversubscription.
edit: Though-
> Let's raise the unit cost to $1.5 million assuming a very good yield on satellites making it to orbit.
How do you get that?! Starlink don't lose 30% of sats, lol.