They're not necessarily using it for what the bottleneck is. They might do raw compute there, but are locked on a DB query or write that's hosted internally or that just is taking awhile to properly partition.