There are lots of possibilities! Not sure if any of them can really achieve “practically unlimited”. Evaluating which one offers the best balance of consistent-low-latency, finality, capacity, practicality, reliability, assurance, simplicity, decentralization, etc. is the next step.
Possibilities include using new ZKP techniques such as zk-STARKS to compress large numbers of transactions (e.g. 1M transactions) into a single proof (e.g. a 1-MB string), and then recursively compress a million such proofs into a 1-MB string, and then have global network consensus on that. This would yield a capacity of up to a trillion transactions per consensus-period. But there are a whole lot of questions about whether it would be possible at all or whether it would come with other trade-offs that are unacceptable.
Other possibilities include Bitcoin-NG or SPECTRE, some new invention that I haven’t understood yet (like Algorand), or “copy whatever Casper/PoS/sharding protocol Ethereum has launched, as long as it works”.