How to find and export your BigQuery data for AI training

01

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

What a BigQuery warehouse records

A data warehouse is a database built to analyze very large volumes of records collected from many of a company's other systems, instead of running one application day to day. BigQuery is Google's warehouse, hosted on Google Cloud. Its tables sit inside datasets, which are folders that group related tables, and each dataset lives in one location, such as the US or the EU.

Companies load into it their website and app events, their sales transactions, their server logs and their advertising data. Many tables are also partitioned by date, which means each day's rows are stored as their own slice, so a long range of dates comes out consistent and each day can be read on its own.

Why linked warehouse tables 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. A warehouse is a strong source because it joins records that no public text contains together. A table of website sessions and a table of purchases can share a customer ID, so each purchase can be traced back to the visits that came before it.

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 histories, such as sessions that ended in a purchase or support contacts that ended in a renewal, into worked examples that an agent studies, or into a practice copy of the business 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 BigQuery fits into a dataset

A BigQuery warehouse 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 BigQuery show how much data you have?

How to read your storage total

The steps below are run by a developer or analyst in the BigQuery query editor, the SQL window in the Google Cloud console. They follow Google's TABLE_STORAGE view reference.

1. Ask a project administrator for the role BigQuery Metadata Viewer (roles/bigquery.metadataViewer) on the project, or for the two permissions bigquery.tables.get and bigquery.tables.list instead. Billing account access is not required, and the view exists in every project, whichever storage billing model it uses.
2. In the query editor, run this statement. It adds up every table in each dataset and shows rows, logical size and physical size side by side:

   SELECT table_schema,
          SUM(total_rows) AS row_count,
          ROUND(SUM(total_logical_bytes)/POW(1024,3),2) AS logical_gb,
          ROUND(SUM(total_physical_bytes)/POW(1024,3),2) AS physical_gb
   FROM `region-us`.INFORMATION_SCHEMA.TABLE_STORAGE
   WHERE NOT deleted
   GROUP BY table_schema
   ORDER BY logical_gb DESC;

   Replace region-us with the location your datasets are in, such as region-eu. The query has to run in that same location, so a company with datasets in several locations runs one query per location and adds the results.
3. For one table only, click it in the Explorer pane and look at Storage info on the Details tab.
4. To learn how far back the data goes, ask your largest tables for their oldest date, for instance SELECT MIN(event_date) FROM mydataset.events;.

Which size columns to add up

The view reports the same tables in several ways, and only one of them is the figure to report.

Why the number can lag or fall short

The view usually updates within minutes, but a change from a table expiring or from a new time travel window can take up to a day to appear. The query also skips deleted tables because of the WHERE NOT deleted filter. A project in a location you forgot to query adds nothing to the total, so check every location that holds a dataset.

03

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

The size in GB

Add up total_logical_bytes across your datasets and write the total in GB. Logical bytes are the uncompressed size, which is the best match for the data you would hand over. Leave out the physical figure, because it is compressed and counts old versions of changed or deleted rows. The query above divides by 1024 three times, so its GB are gibibytes (1,073,741,824 bytes each).

The years of history

Take the years of history from the oldest event or created timestamp inside your main tables. The creation_time column says only when someone created the table, and the records loaded into it may reach back much further.

The number of seats

The seat count is the number of customers or users whose activity the tables record. It is not the number of analysts who run queries in BigQuery.

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 BigQuery history?

What you need first

First create a Cloud Storage bucket that sits in the dataset's own location. The account running the export needs bigquery.tables.export (part of BigQuery Data Viewer) and bigquery.jobs.create, and on the bucket it needs storage.objects.create and storage.objects.delete, both of which come with Storage Object Admin.

Google advises switching off Bucket Lock and Soft Delete retention on the bucket while the export runs. The details are in Google's guide to exporting table data.

The steps, one table at a time

A single extract job exports only one table, so each step below is repeated for every table. You can use any one of the three methods.

1. From the command line, run bq extract once for each table. The * in the file name lets BigQuery write more than one file:

   bq extract --destination_format PARQUET --compression SNAPPY mydataset.mytable gs://my-bucket/mytable/part-*.parquet

2. With SQL, run an EXPORT DATA statement, which also lets you choose the columns and put the rows in order:

   EXPORT DATA OPTIONS (
     uri='gs://my-bucket/mytable/*.parquet',
     format='PARQUET',
     overwrite=true
   ) AS SELECT * FROM mydataset.mytable;

3. In the console, open the Explorer pane, expand the dataset, choose Overview > Tables, select the table and choose More > Export to Google Cloud Storage. In the dialog, pick the bucket location, the export format and the compression, then click Save.

Formats and compression

The console offers only GZIP, so choose bq extract or the SQL statement for Snappy or Zstandard compression.

Limits to plan around

Each file holds at most 1 GB of logical data, and the sizes of the files vary, so a large table becomes many files. On the shared pool of free processing capacity, a project can extract at most 50 TiB a day. Moving the project onto a slot reservation (capacity you pay for) that has the job type PIPELINE removes the cap, and you are then billed at capacity prices.

EXPORT DATA statements are not supported in a PIPELINE reservation, so use bq extract or the console there. A view stores no data of its own, so it can only be exported with EXPORT DATA.

What an export leaves out

An export holds the table as it stands when the job runs, and rows loaded while an extract is running are not included. The order of the rows is not guaranteed unless you use EXPORT DATA with an ORDER BY clause. Time travel and fail-safe copies are not exported, and the file sizes from EXPORT DATA are not guaranteed either.

How long an export takes

None of the Google pages we checked gives a time, because it depends on the size of the table and the capacity available. Because each table is its own job, a project with many tables is exported as a series of jobs, and you can run several at once within the limits described earlier.

Sources: Google Cloud, TABLE_STORAGE view, Export table data to Cloud Storage, Data retention with time travel and fail-safe and Quotas and limits. Steps checked against these pages in November 2026.

05

What stays private when BigQuery data is licensed?

You choose what is in scope

EXPORT DATA runs a query, so you can drop columns such as emails, device IDs or payment fields before anything is written to Cloud Storage. For example, SELECT * EXCEPT (email, device_id) FROM mydataset.events writes every column of the table except those two. The data vendor also deletes records too sensitive to clean safely instead of anonymizing them.

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.