My experience is that MySQL scales as high as you need it when coupled with Vitess, because most data schemas can be sharded, and you can scale each shard vertically enough for all the workloads I've seen.