Seems reasonable, but I'd also stop using most subscriptions before 20 years: 5 years seems more likely. I'd probably cancel a subscription if I stopped using the product after a few years or if the related hardware died.
Well, yes and no. GP is calculating the discounted value of receiving $N annually forever (assuming a fixed constant interest rate r):
N/(1+r) + N/(1+r)^2 + N/(1+r)^3 + ...
= N/r
(You can derive it from (for |q| < 1), 1 + q + q^2 + q^3 + ... = 1/(1-q) )
Note, by the way, that currently rates are very low (say 1%), and thus receiving a perpetual stream of $N per year is very valuable, namely 100N. That makes the switch to subscriptions even more valuable to these psycho firms.
As you say, they're lending you what would've been the upfront payment for the software. They receive your subscription, and have to pay (in interest, to their bank) very low rates. So, on the (fixed) notional they currently pay low rates, while receiving high-ish subscription.
In an ideal market, you would expect the discounted subscription to match the upfront payment, which would mean that the "high value" is simply the price that they have to pay to get those future subscription payments (or that they could sell those future subscription payments for on some sort of "customer subcriptions market" if there were one). Once they've bought them, they also have the risk of rising interest rates wiping out the value of that investment.
(And also, this doesn't really apply all that well in this case anyway, because your typical software license subscription is not a perpetuity, the span of time that people will want to pay for using the software as it is now is rather limited, and they can pretty much stop paying at any point, while you'd supposedly have spent the loan from your bank on your programmers, and your bank will still want to get their interest payments.)