That's also something we've had an awful lot of topics about, heh. 

Ultimately, if Bitcoin had retained the "1 CPU = 1 Vote" concept Satoshi had, scaling would be a far simpler question.  The idea was you were either a miner or an SPV user.  And there was nothing in between.  However, mining became centralised at a rate far quicker than Satoshi had anticipated.  As such, the need for non-mining nodes became essential to keep a reasonable degree of decentralisation.  Notice how there's no mention of non-miners relaying transactions or enforcing consensus rules.  They were never part of the design, but we now rely heavily upon them.  It's because of these nodes that scaling is a far more nuanced and delicate matter.  Unlike miners, there's no financial reward for running a non-mining node, but they do help to secure the network.  Ergo, making it more increasingly more costly to run something that earns no rewards is a difficult ask.