The medium article has been refuted quite a few times in the comments section. I wouldn't read much into the article because of the assumptions made there. The writer has changed his position in subsequent comments to "It won't work because the central entities would be too large". This is again a big assumption. Any type of payment channels can be established between parties depending on the amount of funds they move. Some will be large, some will be small.
They make some excellent points. In short, I see LN fail because it's a really dumb system. Users have to be online 24/7 to form part of a payment route, the user with the lowest amount of money determines how much you can pay in that route (big LOL!), extreme data transfers are needed because the network needs to know the state of everybody's payment channel all of the time in order to discover routes, wiring BTC into/out of your payment channel still is an on-chain transaction, users can defraud each other in a payment channel, no guarantee you can pay the person you'd like to pay etc etc etc
You are trying to find reasons that why it'd fail so you are under-estimating the capabilities of a complex network, when fully deployed. LN is supposed to solve the problem of micro transactions particularly. These transactions are never going to be worth more than a fraction of a BTC (at present values).
The number of people who hold these amounts are not trivial. The number of people who would be willing to act as a hub will determine the success of the network. If you have enough people ready to commit part of their holdings to a payment channel for a fixed time, you will have sufficient routes available for any transaction. For the large transactions, there already is on-chain settlement.
I can't wait to see it in action and am going to look for reasons as to why it will work, rather than why it won't because clearly, there are no other ideas we have seen implemented or tested for scaling as yet.
Also, its quite easy to point out why a complex system won't work.
Turbo-generators weigh multiple tons but they still rotate at mindblowiing speeds of 3000-3600 RPM suspended on a layer of oil. By all accounts, it seems intuitively pretty tough.
Any complex man-made system doesn't seem to work intuitively until it actually does.