Earlier quoted context omitted.
> So doing even 1000/sec means around 1.8Mbps (plus overhead) just to get the flow of transactions. Huh, maybe we should come up with a cryptocurrency where you are awarded new blocks for the storage and bandwidth you contribute to the effort, rather than for how many hashes you can calculate per second. After all, if we're hoping to handling transactions on the scale of Visa, and we don't like centralization, we nee…
They don't need to store everything. All that's needed is the list of addresses and balances (UTXOs), and the last few blocks. A hash chain of 10 years of 10 minute blocks is only a few MB, so it should be something we can distribute reliably. UTXOs are a couple GB but will grow.
(I'm not personally sold that this is actually a distinction that matters - after all, we are already implicitly trusting "authoritative nodes" to deliver the blockchain, and the same mechanism that prevents alteration of the blockchain right now would also prevent altering the database state. Namely, that nobody else is going to accept the version of history where I own infinity bitcoins, regardless of how we record that state.)
A middle ground might be to have the full state available to nodes that want to verify, but also implement a "checkpoint/epoch" mechanism on top for "fast clients" who aren't overly concerned about a massive concerted-but-silent attack on Bitcoin escaping notice.