This seems to be a common view, but I'm not sure why.
Bitcoin is designed to be able to scale to very high levels because it is able to prune older transactions. Full nodes only need the header chain and UTXO set to operate (plus some amount of full block recent history). If Bitcoin scales to a very large scale I suspect we would see most full nodes do not maintain the full history (because they don't have to) and there would also exist a few archival nodes run as public services that contain the full history. This can all be done on the main chain as is today.
In the end bitcoin will scale to whatever the marginal cost of adding a transaction is vs the increased orphan rate risk. With IBLT and sub-chains this calculation works out to a very low fee per transaction which means bitcoin can and will accept coffee level transaction simply because it will be profitable to do so.
Storage is not a concern, they keep developing new tech for cold storage. Engadget just had an article on a new nano dot method that puts 350TB on a 1 inch square object. A few publicly funded archival nodes could contain a lot of coffee transactions.....
It's about visions of bitcoin's final state.
We already see Moore's Laws fading. E. G. Supercomputers did nearly stagnate the last 3-4 years. It's a mixture of physical limits, the satisfaction of needs and a trade-off between energy consumption und benefits. More important than enhancing home computers is it to enhance smartphones, that's why the development of home cpus fades. Supercomputers fullfill the needs of most scientific needs, the benefit of more giga/petaflops are limited, while the energy consumption grows exponentially. My home internet connection - it's really slow, 20/2 mbs - is enough to satisfy my needs and that of all my neighbours - streaming videos, uploading images, listening to music - there's not really a need to enhance it.
So ... IF technological progress can't keep up with Bitcoin-system eating the world's transactions, we will end up with a state, where private users are unable to connect directly with the network. If connecting to bitcoin means to first download and verify one day of 2GB-blogs and then being a node in an ocean of 2gb-blocks, than it will be too much for most home users. It's already not pleasure, but it's possible, and I highly prefer to do it and I think its ideologically essential to be able to do so.
Is this a problem? Not directly. Maybe we will have 10-20 Nodes, and every users pays a little, little fee to process his transactions. This nodes will be the new banks, but that state is much better as the current states, cause the users will be able to store their private key by themself (or by multisig) and they maybe have the choice to write a transaction by themself and give it to any node.
But it's not the perfect scenario for several reasons. It brings back the middle man and it adds some attacks by man with gun knocking on doors. Better would be a scenario where home users still can connect to the network without an intermediare while the network handles millions of transactions each second.
But it remains a bad idea to damage the current network. We should scale onchain as far as technologie allows the network to remain decentral.
To be honest, I think all I've written is fantasy, because Bitcoin will never ever be the favorite method to pay your coffee. Even I, as a bitcoin fanboy, prefer to pay my coffee and my beer with cash, because it's usually more comfortable. The most possible scenario is that bitcoin will never reach a number of transaction that can't be processed onchain.
P.S.: when nodes don't safe the whole blockchain we risk to loose the function of the bitcoin blockchain as a supernotary for any kind of information. I don't think this is a good idea. But maybe I just have a lack of knowledge about pruning.