You can find some old forum posts going in this direction by searching for “execution engine” or “state minimization”. e.g.: State-minimised executions Delayed state execution, finality and cross-chain operations. Note that sometimes the posts are tightly coupled with sharding research.
The common starting point is the observation that any execution (including smart contracts) can be simulated on top of a “data-only” blockchain.
Some challenges I see with it though:
- Without a native token, there is no in-protocol incentivization for validators and block producers. The idea of paying through a token whose balance is tracked by
compute_dapp_stateis similar to economic abstraction (https://medium.com/@Vlad_Zamfir/against-economic-abstraction-e27f4cbba5a7) - Requirement for a user to download all historical data, which might be much larger than the “current state” (e.g. UTXO set vs TXO set) as @hkalodner mentioned
- At the protocol level, all historical data must be stored somewhere and the protocol “fails” if some historical state is completely deleted (stored by no one). In “stateful” protocols there is a smaller set of current state that nodes are required to store, we can delete things from the current state and incentivize/force people to do so e.g. via charging rent, hence controlling the storage requirements placed upon nodes.