01
Storage and compute scale apart
A petabyte sitting in object storage costs the same whether one query touches it or one thousand do — the warehouse spins compute only when work arrives.
Idle storage cost · constant
02
Virtual clusters never share a queue
Finance, marketing, and the data-science team each run on an isolated compute slice — a heavy load on one slice does not slow down the next.
Median query wait · under 1.4s
03
Data lives once, in one cold lake
A single immutable copy is fanned out to every region through metadata pointers — no nightly replication job, no drift between marts, no Sunday-morning rebuild.
Cross-region copy time · zero
04
Time travel is built into the table
Every change leaves a versioned snapshot, queryable for ninety days — restoring a dropped column or auditing yesterday's number is a single SELECT, not a ticket.
Recovery window · 90 days