[vmception]

You need an archival node to store all the states and also monitor unconfirmed transactions. The primary client go-ethereum (geth) creates archival nodes that grow 1 terabyte per year (7tb now).

Despite archival nodes not being needed for consensus, some people’s kneejerk response is that this means Ethereum is not sustainable, but really geth just has unoptimal coding.

There are other clients out there that have chosen different data structures and reduced this by an order of magnitude. Turbo-geth, for example can sync an archival nodes at 1tb and is working on other improvements.

There is more support for geth at the moment but this is also an area of development if you want to improve upon.

Its likely many nodes out there are using their own optimizations and are not offered to the community.