Because the Algorand protocol uses pure proof-of-stake to randomly select a small set of block proposers and verifiers, users only need to receive a fixed number of messages in order to reach consensus on the next block. This is in contrast with previous Byzantine Agreement protocols, which either require every user to receive a message from all other users, or rely on ad-hoc chains of trust.
Unlike proof-of-work protocols like Bitcoin, Algorand’s consensus protocol does not require participants to solve cryptographic puzzles in order to propose or validate blocks. In Bitcoin, this computation is simply wasted work, and if more computational power is directed toward mining, the underlying puzzles must be made more difficult to maintain the desired block generation rate, thus increasing computational waste without improving the performance of the blockchain.
In contrast, in the Algorand protocol, the computation cost a single user faces only involves generating and verifying signatures and simple counting operations. The cost depends only on the number of selected users for each block, which is constant and not affected by the size of the whole network. If computation ever becomes a bottleneck for a user, then increasing the computational power will directly improve performance.
As a result, Algorand’s consensus protocol is able to scale to millions of users and sustain a high transaction rate, without incurring significant cost to participating users. Consensus on a block is reached in parallel while the block is being propagated to the network, which typically happens in a few seconds.