A slot is a unit of compute — roughly one virtual CPU plus some memory — that executes one piece of a query stage at a time. Slots are owned by Google and live in a shared pool inside each region. They're not VMs you provision; they're abstract scheduling units the Dremel scheduler hands out per millisecond. A query doesn't get "a slot" — it gets a time-varying allocation across the lifetime of its stages. A query might use 2,000 slots for 800 ms during a wide scan, then 50 slots for 200 ms during final aggregation, then release them. The scheduler is constantly reassigning slots between queries at sub-second granularity.