The problem is that the initial sync slows to a crawl when UTXOs are stored on disk rather than in RAM.

You can see this effect most clearly in low-resource devices like Raspberry Pis with 4GB of RAM, which used to be able to sync the chain in just a few days, but now take around a month. This has resulted in these devices becoming practically useless as bitcoin nodes, whereas a year ago (when the utxoset was half as large as it is now), they were usable.

I consider Raspberry Pis to be the canary in the coalmine wrt UTXO set growth. Although they should probably never have been recommended, we should consider it a failure as a community when a whole class of devices is excluded from participating in the bitcoin network.

We need to be making it *easier*, not harder, to run a full node, and I consider this to be the #1 priority of the bitcoin community, as it is already extremely rare to find merchants who run their own nodes.

Although most computers can fit 10GB in memory at the moment, if the UTXO set were to expand to 50GB, which it could theoretically do in just a year in an attack scenario, the majority of computers would be unable to fit the UTXO set in memory, and would thus be less and less likely to be able to sync bitcoin in a reasonable amount of time.

Obviously 400GB does not fit into RAM unless you are spending tons of money on high-end servers. Let's avoid requiring high-end servers to use bitcoin.