@markmc, Ethereum and Algorand were designed with very different approaches, and hence ended up with very different solutions.
Ethereum virtual machine is executing EVM bytecode, as long as you’ve provided the proper amount of gas ( fee ). So, while it’s true that it’s turing complete, some types of programs could be inherently too expensive to be executed. Current Ethereum gas price are a proof for that.
Algorand took a very different approach. The core contract language, TEAL, is not turing complete. Nevertheless, it still allow you to implement (almost) any computation needed for DeFi applications. That allows Algorand to operate at a high speed, providing low-latency L1 contracting solution. For few solutions, look here - Algorand Developer Portal.
That L1 solution would meet the needs of 99% of the users / applications. For the reminder, there are two different approaches that are being explored ( no official products around these yet ) : one is to create a co-chain that would exchange tokens with the mainnet. This would allow the co-chain to run slower than the mainnet, and perform heavy computations. Another solution is to implement a layer 2 (L2) application that would interact with some L1 primitives to ensure operational atomicity.
Last, there are companies that attempts to bridge the gap by providing a smooth transition between building for Ethereum vs. building for Algorand. See Reach - https://algorand.foundation/news/reach-development