How to find and export your Snowflake data for AI training

01

What is Snowflake data, and why do AI labs and data vendors buy it?

What a Snowflake account records

A data warehouse is a database built to analyze large volumes of records copied in from a company's other systems. Snowflake is one such warehouse, run as a cloud service. Its tables sit in databases, and each database is divided into schemas, which are folders that group related tables.

Snowflake is often the place where a company's systems meet. Events from the product, invoices from the billing system, deals from the CRM (customer relationship management software, where a sales team keeps its customers and deals) and tickets from the support desk land side by side. Data teams then clean them into modeled tables and match the customer IDs between sources.

Why matched records teach an AI model something

An AI model learns by studying a very large number of examples and finding the patterns in them, and those examples are called training data. Warehouse tables hold years of history at the level of single rows, each with a consistent timestamp. A model can read what a customer did, what the company did in response and what happened next, which public text on the internet almost never shows.

AI labs now want to train AI agents, software that carries out a job step by step, and an agent learns a job from records of that job being done. A data vendor can turn anonymized, linked records into worked examples that an agent studies, or into a practice copy of a business process in which the agent is scored on each attempt. The full mechanism is in AI training data: why AI companies want your company's data.

How Snowflake fits into a dataset

A Snowflake account is usually one of several sources in a company's dataset, next to email, documents and tickets. Companies are paid $100K to $2M or more per dataset, according to micro1, the data vendor we typically refer companies to. The price covers the whole dataset, not each source on its own, and what your data is worth explains what moves it.

02

Where does Snowflake show how much data you have?

How to read your storage total

The steps below are run by a Snowflake administrator or a data engineer. They follow the TABLE_STORAGE_METRICS reference and the guide to exploring storage cost.

1. Sign in with the ACCOUNTADMIN role. Any other role needs an administrator to run GRANT DATABASE ROLE SNOWFLAKE.USAGE_VIEWER TO ROLE your_role; first. The ACCOUNT_USAGE views are open to ACCOUNTADMIN only by default, and a role can also read them through IMPORTED PRIVILEGES on the SNOWFLAKE database. Every Snowflake edition has these views.
2. To see storage over time in Snowsight, open Admin > Cost management, pick a warehouse (Snowflake recommends an X-Small one), choose Consumption and set Usage Type to Storage. Snowflake warns that cost data may take as long as 72 hours to show up in Snowsight.
3. For the size of current table data, run this query in a worksheet. It adds up every table that has not been dropped and prints the three kinds of storage in separate columns:

   SELECT table_catalog,
          ROUND(SUM(active_bytes)/POW(1024,3),2)      AS active_gb,
          ROUND(SUM(time_travel_bytes)/POW(1024,3),2) AS time_travel_gb,
          ROUND(SUM(failsafe_bytes)/POW(1024,3),2)    AS failsafe_gb
   FROM SNOWFLAKE.ACCOUNT_USAGE.TABLE_STORAGE_METRICS
   WHERE deleted = FALSE
   GROUP BY table_catalog
   ORDER BY active_gb DESC;

4. For a daily figure per database, run the query below. Its database_gb column includes data held for Time Travel, so it reads larger than the active figure:

   SELECT usage_date, database_name,
          ROUND(average_database_bytes/POW(1024,3),2) AS database_gb,
          ROUND(average_failsafe_bytes/POW(1024,3),2) AS failsafe_gb
   FROM SNOWFLAKE.ACCOUNT_USAGE.DATABASE_STORAGE_USAGE_HISTORY
   WHERE deleted IS NULL
   ORDER BY usage_date DESC, database_gb DESC;

5. For a single table, open Catalog > Explorer, pick the database and then Tables. The size shown there is the table's active bytes.
6. To find the first year, query the oldest business date in your main tables, such as the earliest order date or signup date.

Which storage columns to add up

Snowflake keeps several kinds of storage for the same tables, and only the first is the figure to report.

Why the number can lag or fall short

The table view can run up to 90 minutes behind, and the daily per-database view up to 3 hours. Both views hold just the last 365 days, so neither can tell you when loading began. Snowflake also notes that database totals will not equal the storage on your bill, because the bill adds staged files and other items.

03

What should you know about your Snowflake data before you apply?

The size in GB

Add up ACTIVE_BYTES across your tables and write the total in GB. Snowflake compresses all table data and measures storage by the compressed size, so the same tables written out as uncompressed CSV take up more space. Don't add the Time Travel, Fail-safe or clone figures, because they are billed storage for changed, deleted or copied data.

The years of history

Read the years of history from the oldest business date in your main tables, for instance the first order. The ACCOUNT_USAGE views are no help here, because they retain a single year and so never reveal when loading began.

The number of seats

The seat count is the number of customers or users whose activity the tables record. Snowflake logins, the accounts your own staff use to run queries, are not seats.

What you don't need to do first

You don't need to export anything to apply, and you don't need to clean or scrub the records. The data vendor anonymizes them before anyone else sees them, and nothing moves until a lawyer-reviewed license is signed. You only need the size, the years and the count for the first conversation. The application, the intro call and each later step are laid out in how to sell your company's data.

04

How do you export your full Snowflake history?

The steps with COPY INTO

Each step is run in a worksheet or in SnowSQL, Snowflake's command-line tool. Snowflake describes the command in its COPY INTO location reference.

1. Create a stage where the files will land. A stage inside Snowflake (an internal stage) is enough for a moderate amount of data:

   CREATE STAGE export_stage;

   For a large amount of data, create an external stage pointing at your own Amazon S3, Google Cloud Storage or Azure container, secured by a storage integration. That is a Snowflake object holding a trusted link to your cloud account, so no keys are typed into commands.
2. Unload each table to Parquet. Snowflake writes several files in parallel:

   COPY INTO @export_stage/orders/
   FROM mydb.public.orders
   FILE_FORMAT = (TYPE = PARQUET)
   HEADER = TRUE
   MAX_FILE_SIZE = 268435456;

3. If you want CSV instead, set FILE_FORMAT = (TYPE = CSV) HEADER = TRUE. CSV output is compressed with gzip by default.
4. Download the files from an internal stage with the GET command in SnowSQL or the Snowflake CLI (snow sql):

   GET @export_stage/orders/ file:///data/export/;

   Snowflake does not let GET run inside a worksheet in either web interface. After an unload to an external stage, no download is needed because the files already sit in your bucket.

Writing directly to a bucket in S3, Google Cloud Storage or Azure requires a storage integration or access credentials. Each unload also needs a running virtual warehouse, which means it spends credits.

Formats and limits

The default MAX_FILE_SIZE is 16 MB, and you can raise it to 5 GB, though it works as a target and not a strict ceiling. Add SINGLE = TRUE when you would rather have one file than many.

What an export includes and leaves out

An unload holds the table as it stands when the statement runs. If you need an earlier state, Time Travel data can be queried with AT or BEFORE clauses, whereas Fail-safe copies are out of your reach: you cannot query, clone or restore them.

How long an export takes

Snowflake publishes no time, because it depends on table size, the size of the warehouse and the number of files. A larger warehouse unloads faster and uses more credits.

Sources: Snowflake documentation, TABLE_STORAGE_METRICS view, Understanding storage cost, COPY INTO location, Understanding and using Time Travel and GET. Steps checked against these pages in November 2026.

05

What stays private when Snowflake data is licensed?

You choose what is in scope

COPY INTO accepts a SELECT in place of a table name, which lets identifiers, payment fields and free-text notes stay inside Snowflake while the other columns leave. This statement, for example, writes only three columns of one table:

COPY INTO @export_stage/orders/
FROM (SELECT order_id, created_at, status FROM mydb.public.orders)
FILE_FORMAT = (TYPE = PARQUET)
HEADER = TRUE;

Names are replaced before a lab sees them

Anonymization means replacing every detail that identifies a person or a company, such as a name or an email address, with a stand-in that is the same every time it appears. The records still link to each other and the dates still run in order, so a lab can learn from them, but they no longer say who anyone is. The diagram shows one set of records before and after.

The method, including what is deleted outright and how you review samples, is in how company data is anonymized for AI training.

Ownership and payment

Licensing gives the buyer limited rights to a copy, while the company keeps ownership of the originals. The terms that matter, such as exclusivity, permitted use and deletion, are covered in data licensing. micro1 pays in full within 30 days of accepting the data, usually as a single payment.