>
Pruned nodes are not the default configuration, if it was the default configuration then I think you would see far more users running a pruned node.
Default configurations aren't a big enough deal to factor into the critical discussion of node costs versus transaction fee cost. Default configurations can be changed, and if nodes are negatively affected by a default configuration, there will be an abundance of information about how to correct that effect by turning on pruning. Bitcoin can't design with the assumption that people can't google - If we wanted to cater to that population group right now, we'd need 100x the blocksize at least.
> But that would also substantially increase the burden on archive nodes.
This is already a big problem from the measurements I've been looking at. There are alternatives that need to be considered there as well. If we limit ourselves to not changing the syncing process for most users, the blocksize limit debate changes drastically. Hard drive costs, CPU costs, propagation times... none of those things matter because the cost of sync bandwidth is so incredibly high even now ($130ish per month, see other email). Even if we didn't increase the blocksize any more than segwit, we're already seeing sync costs being shifted onto fewer nodes - I.e., Luke Jr's scan finding ~50k nodes online but only 7k of those show up on sites like
bitnodes.21.co. Segwit will shift it further until the few nodes providing sync limit speeds and/or max out on connections, providing no fully-sync'd nodes for a new node to connect to. Then wallet providers / node software will offer a solution - A bundled utxo checkpoint that removes the need to sync. This slightly increases centralization, and increases centralization more if core were to adopt the same approach.
The advantage would be tremendous for such a simple solution - Node costs would drop by a full order of magnitude for full nodes even today, more when archival nodes are more restricted, history is bigger, and segwit blocksizes are in effect, and then blocksizes could be safely increased by nearly the same order of magnitude, increasing the utility of bitcoin and the number of people that can effectively use it.
Another, much more complicated option is for the node sync process to function like a tor network. A very small number of seed nodes could send data on to only other nodes with the highest bandwidth available(and good retention policy, i.e. not tightly pruning as they sync), who then spread it out further and so on. That's complicated though, because as far as I know the syncing process today has no ability to exchange a selfish syncing node for a high performing syncing node. I'm not even sure - will a syncing node opt to sync from a different node that, itself, isn't fully sync'd but is farther ahead?
At any rate, syncing bandwidth usage is a critical problem for future growth and is solvable. The upsides of fixing it are huge, though.