The just-rent-servers scheme presents the following issues:

1) Won’t get you anywhere near as good a network of globally-distributed servers. [edit] I mean, maybe, but at that point you’re looking at a fair amount of cost in vendor management—you’ll probably need several hosting providers to even approach it.

2) You need minimum three good full-time ops people to build and support such a thing. What’s the salary on that over two years? Even with Polish salaries or what have you, it’s cheaper, but not cheap. Let’s say $300,000 fully-loaded cost over two years, just for a nice round number, for three ops/sysadmin sorts. Even in cheaper markets, that may be underestimating what it’d take. Halfway there already, and you haven’t rented servers or rack space yet. I’m not gonna claim that your infra will also work far less well and break more often than Cloudflare, because sometimes these sorts of set-ups do end up being rather stable because they can be far simpler and smaller-scale than an as-a-service product, but that is a risk if it’s not done quite well.

3) Server providers may drop you if you get too expensive. Get DDOS’d too much, just use their bandwidth too heavily but within your nominal limits. It’s a risk.

4) Those rented servers do cost money, and add to the operations salaries costs above. Buy servers? Also money, and now you need to rent rack space and bandwidth.

5) Peering agreements for server hosts are sometimes bad. It’s hard to know whether yours is bad until you do something at-scale on it and see support tickets roll in for consistently dial-up speeds for clients who really ought to be seeing better than that. This info doesn’t make service comparison spec sheets.

6) CDNs have usually put in some effort to solve e.g. China deliverability. You’re starting from scratch as far as political difficulties go.