Definitely yes to the first: the index implementation is fully integrated with PlanetScale's sharding layer, so you can scale horizontally as much as you need. This works very well in practice: the sharding layer, Vitess, is extremely well battle tested with petabyte-sized clusters in production, and our vector indexes are like any other table on the cluster, so it really scales horizontally with very predictable performance.
As for separating storage and compute: we don't do that. One of our key sells here is that this is vector data fully integrated into your relational database with ACID semantics. Very hard to do separate storage and compute and keep this behavior!
As for separating storage and compute: we don't do that. One of our key sells here is that this is vector data fully integrated into your relational database with ACID semantics. Very hard to do separate storage and compute and keep this behavior!