How to Use Snowflake Time Travel to Recover Deleted Data
A bad DELETE can be undone in Snowflake without a backup โ if you act inside the retention window. Here is the exact recovery sequence…
Snowflake is a cloud data warehouse where storage and computing are billed separately. You load data once, then spin up warehouses of different sizes against it – and because compute bills per second, a query that runs for 40 seconds costs 40 seconds.
It is a SQL data platform that runs on AWS, Azure or Google Cloud but manages its own storage layer, so you are not locked to one cloud’s file format. Analysts query with normal SQL; the engine handles concurrency by assigning separate compute clusters per workload.
Snowflake went public in 2020 and reported more than 12,000 customers by 2025. Cortex, its AI layer, adds functions like SENTIMENT, SUMMARIZE and COMPLETE that run inside SQL – which means an analyst can summarize 50,000 support tickets without leaving a worksheet.
Credits, not dollars. Compute is priced per credit, roughly $2 to $4 depending on cloud and region; an X-Small warehouse burns one credit per hour, so an hour of small-warehouse compute sits near $3. Storage runs about $23 per terabyte per month on-demand, less with upfront capacity.
The failure mode is well documented: an oversized warehouse left running, or a dashboard that refreshes every five minutes against a Large warehouse, can burn thousands a month. Set resource monitors and statement timeouts on day one. Cortex LLM calls bill per token on top of compute.
Best for companies past the point where a single Postgres instance chokes – typically 50GB to petabytes, with a mix of BI dashboards, data science and ad-hoc analysis hitting the same tables. Mid-market and enterprise teams with a data engineer on staff get the most out of it.
Not for a 20-person startup with 2GB of transactional data; a managed Postgres or BigQuery free tier is cheaper and simpler. Also a poor fit if your team has zero SQL skill – nothing here replaces an analyst.