Not everyone running a node has a plethora of available resources for it. It would be great if we could keep scaling bitcoin by just increasing the blocksize, but by the time you get to a block size like 8MB or 32MB, things get really difficult.
From a Bitfury paper on the subject [1]:
8MB blocksize requires: 411GB/year extra space, 1184kB/s average bandwidth, 35.4TB/year bandwidth, and 32GB RAM.
If that isn't enough to scale, which it won't be if Bitcoin is to fulfill its goals of being able to compete with the currently leading payment networks, take a look at something like 32MB blocks:
32MB blocksize requires: 1643GB/year extra space, 4736kB/s average bandwidth, 141TB/year bandwidth, and 128GB RAM.
Although these numbers are possible to meet, it's obvious that not everyone is going to have hardware like this just to run a Bitcoin node. So it's predicted that so many nodes will be excluded for larger blocksizes (90%+), that the decentralization will take a huge blow.
And what kind of blow will decentralization take when no one wants to run a node that only serves the purpose of settlement for a few large players?
The BW calculations are misleading - that counts transmitting the blockchain to other nodes. To validate transactions and participate, you don't need to upload it, only download. Also, if we consider transaction history irrelevant since Lightning or other offchain solutions are OK, then all that is needed is the last few blocks and UTXO.
As Satoshi said, once client-only is in, it doesn't really matter, and the answer is to not care how big it gets. (Sure, quoting him isn't an end-all, but it shows the original design goals.)
Also, none of this explains why an increase to say, 2MB was never done, unless it's a game of chicken.
From a Bitfury paper on the subject [1]:
8MB blocksize requires: 411GB/year extra space, 1184kB/s average bandwidth, 35.4TB/year bandwidth, and 32GB RAM.
If that isn't enough to scale, which it won't be if Bitcoin is to fulfill its goals of being able to compete with the currently leading payment networks, take a look at something like 32MB blocks:
32MB blocksize requires: 1643GB/year extra space, 4736kB/s average bandwidth, 141TB/year bandwidth, and 128GB RAM.
Although these numbers are possible to meet, it's obvious that not everyone is going to have hardware like this just to run a Bitcoin node. So it's predicted that so many nodes will be excluded for larger blocksizes (90%+), that the decentralization will take a huge blow.
[1] http://bitfury.com/content/5-white-papers-research/block-siz..., page 4