That is exactly what Ethereum's direction has been for years now. It is centralized on Amazon AWS via Infura's nodes (which they charge access to, mind you). It's not a world computer. It's EC2 with additional complexity (and fees).
The delegation of node operation to third party services is done more for convenience/up-time guarantees than any node operation cost considerations, and Ethereum would be fine if these third party services were all coopted or forced to shut down, because again, Ethereum's consensus protocol has zero dependencies on them.
BTW history-sharding[1] isn't that complicated. If a DLT is build from the ground up with payment in mind and thus history is completely optional then you dont have to do any tech magic. But ofc this is not the case for Ethereum as it was not made for payment.
But you conclude with your worry that there are real costs, and zero pitch decks or white papers I've seen include arguably the most important competitive factor: cost of "how people do it now" vs. cost doing it "this new way;" and in-person, real life networks of trust networks has been how we've become as successful as we have as a society - and so works quite well - and doesn't inherently lineup with "trustless" propaganda of these popular blockchains; if blockchain is a valuable tool, say for legal institutions, they could create their own and run it themselves under agreement.
Mimblewimble[0] is one interesting solution to this. Unfortunately, the requirement in current implementations (such as Grin) that both wallets be online to complete a transaction, eliminates some valuable types of transaction, such as sending coin to a cold wallet.
I do think for a 'world computer' like Ethereum, being able to ignore a substantial amount of old state is going to be critical for long-term use.
I also think that "blatant lie" is unnecessarily harsh. A blockchain is aspirationally forever, and it is a massively replicated data structure: it's certainly durable, and I would expect it to last a long time relative to, say, a random torrent.
This is my major concern with the direction Ethereum is going, which I could caricature as "scale up massively and keep everything forever": it becomes so expensive to keep copies of all data, that only large institutional players will bother, and not many of them. This makes blockchain integrity a "take my word for it" kind of thing, and the whole system is only as durable as the increasingly-enormous data centers which hold all that information.
[0]: https://scalingbitcoin.org/papers/mimblewimble.txt