BigQuery logical vs physical storage billing: 10 TiB cost example

BigQuery bills dataset storage in one of two ways. Logical billing, the default, charges for the uncompressed size of your tables, while physical billing charges for the compressed bytes actually kept on disk at a higher rate per GiB. As listed on Google Cloud's BigQuery pricing page on November 22, 2026, active physical storage costs about 1.74 times as much per GiB as active logical storage, and long-term physical storage about 1.25 times as much as long-term logical storage. Physical billing therefore saves money only when your data compresses well enough to more than cover that gap.
There is a second factor that decides the outcome. Under physical billing, the bytes held for time travel and fail-safe are billed at the active physical rate, while under logical billing they are free. A dataset whose tables are rewritten every day can carry a large volume of time travel and fail-safe bytes, and that overhead can wipe out the compression savings entirely.
The worked example below uses an illustrative 10 TiB dataset to show both effects with real list prices. For a 10 TiB dataset with a 3:1 compression ratio and little churn, physical billing cuts the monthly storage bill from about $192.51 to about $103.57; with heavy daily rewrites, the same dataset costs more under physical billing than under logical billing.
··········
BIGQUERY STORAGE PRICES FOR LOGICAL AND PHYSICAL BYTES.
Google Cloud publishes BigQuery storage prices per GiB-hour. The table shows the rates listed for the US multi-region and the US regions on the pricing page on November 22, 2026. The monthly column is a conversion computed for a 730-hour month, so the figures are comparable with older per-GiB-month price lists; Google Cloud bills storage prorated per MiB, per second.
........
Storage type | Published price | About per GiB, 730-hour month |
Active logical | $0.000031507 per GiB-hour | $0.023 |
Long-term logical | $0.000021918 per GiB-hour | $0.016 |
Active physical | $0.000054795 per GiB-hour | $0.040 |
Long-term physical | $0.000027397 per GiB-hour | $0.020 |
........
The first 10 GiB of storage each month is free under the BigQuery free tier, which also covers BigQuery ML models and training data stored in BigQuery. Google Cloud defines a GiB as 2^30 bytes, and it measures usage in GiB-months.
Active storage is any table or partition modified in the last 90 days. A table or partition that has not been modified for 90 consecutive days moves to long-term storage, and its price drops by approximately 50% automatically. Google Cloud states there is no difference in performance, durability or availability between the two. Each partition of a partitioned table is evaluated on its own, so a date-partitioned fact table usually has most of its history in long-term storage and only recent partitions in active storage.
The 90-day timer restarts when the table is edited. Loading data, copying into the table, writing query results to it, running DML, CREATE OR REPLACE TABLE and streaming through the Storage Write API all count as modifications. Querying the table, creating a view on it, exporting it, copying it to another destination and patching the table resource do not reset the timer.
··········
HOW LOGICAL AND PHYSICAL BILLING TREAT TIME TRAVEL AND FAIL-SAFE.
Time travel lets you query or restore data as it was at an earlier point within a window that is seven days by default and can be configured from two to seven days. After the time travel window, BigQuery keeps the data for another seven days in fail-safe storage; that period cannot be configured or extended, and recovering fail-safe data requires Cloud Customer Care.
The pricing page states the billing rule plainly: there are no charges for time travel and fail-safe storage under logical storage, and those charges apply under physical storage. The datasets documentation adds that, with physical billing, time travel and fail-safe bytes are charged separately at active storage rates.
........
Rule | Logical billing | Physical billing |
Bytes measured | Uncompressed | Compressed on disk |
Default for new datasets | Yes | No |
Time travel bytes | Not charged | Active physical rate |
Fail-safe bytes | Not charged | Active physical rate |
Wait before switching again | 14 days | 14 days |
Time for a change to apply | 24 hours | 24 hours |
........
The billing model is a dataset property. According to the BigQuery documentation (last updated November 10, 2026), a dataset created without a storage billing model uses logical billing, a change takes 24 hours to take effect, and after a change you must wait 14 days before changing the model again. Tables and partitions already in long-term storage are not reset to active storage by the switch, and query performance is not affected.
One restriction applies: a dataset cannot be enrolled in physical billing if the organization has legacy flat-rate slot commitments in the same region. Commitments purchased with a BigQuery edition do not block the switch. For how compute charges work alongside storage, see the comparison of BigQuery on-demand per-TiB pricing and editions slot-hours.
··········
BIGQUERY STORAGE COST FOR A 10 TIB DATASET: WORKED EXAMPLE.
The example is illustrative. A BI team stores 10,240 GiB (10 TiB) of uncompressed data in one dataset: 4,096 GiB in active storage and 6,144 GiB in long-term storage, mostly old partitions of event and order tables. The free 10 GiB per month is ignored because it changes the result by about $0.23. Prices are the monthly equivalents from the first table.
Under logical billing the monthly bill is 4,096 GiB at $0.023 plus 6,144 GiB at $0.016, or about $94.21 plus $98.30, for a total of about $192.51. Under physical billing, the bill depends on the compression ratio and on the time travel and fail-safe bytes. Two churn profiles are compared: low churn, with about 100 GiB of compressed time travel bytes and 100 GiB of fail-safe bytes, and high churn, where roughly 200 GiB of compressed data is rewritten every day and seven days of old versions sit in each of the time travel and fail-safe windows (about 1,400 GiB each).
........
Compression | Churn | Physical monthly | Logical monthly | Physical saves |
2:1 | Low | $151.36 | $192.51 | $41.15 |
3:1 | Low | $103.57 | $192.51 | $88.94 |
4:1 | Low | $79.68 | $192.51 | $112.83 |
2:1 | High | $255.36 | $192.51 | -$62.85 |
3:1 | High | $207.57 | $192.51 | -$15.06 |
4:1 | High | $183.68 | $192.51 | $8.83 |
........
At low churn, physical billing becomes cheaper once the compression ratio passes about 1.55:1 for this mix of active and long-term data. At high churn, the break-even ratio rises to about 3.56:1, because 2,800 GiB of time travel and fail-safe bytes are charged at about $0.040 per GiB each month.
Shortening the time travel window narrows the gap. If the high-churn dataset at 3:1 compression keeps only two days of time travel (about 400 GiB) while fail-safe stays at seven days, the physical bill drops to about $167.57 and physical billing saves about $24.94 a month. At 4:1, the saving under the same setting is about $48.83. The trade-off is a shorter self-service recovery window, which some teams cannot accept for production tables.
Compression ratios for BigQuery data are specific to each table, which is why the vendor provides a forecasting method instead of a single rule of thumb. The same reasoning appears in the breakdown of Snowflake credits, editions and storage pricing, where storage is billed on compressed bytes as well.
··········
FORECASTING THE BILLING MODEL WITH INFORMATION_SCHEMA.TABLE_STORAGE.
BigQuery exposes the numbers needed for the decision in the INFORMATION_SCHEMA.TABLE_STORAGE view. Its columns include active_logical_bytes, long_term_logical_bytes, active_physical_bytes (which includes time travel bytes), long_term_physical_bytes, time_travel_physical_bytes and fail_safe_physical_bytes; fail-safe bytes are not included in total_physical_bytes. Reading the view requires bigquery.tables.get and bigquery.tables.list, which the BigQuery Metadata Viewer role includes.
The documented method prices logical datasets with active and long-term logical bytes, and physical datasets with active physical bytes plus fail-safe bytes at the active physical rate, plus long-term physical bytes at the long-term rate. The query below follows that method with the monthly equivalents of the November 22, 2026 list prices for US locations. The official example on the documentation page uses older default prices of 0.02, 0.01, 0.04 and 0.02 for us-central1, so replace them with the rates for your location before trusting the result.
DECLARE active_logical_gib_price FLOAT64 DEFAULT 0.023;
DECLARE long_term_logical_gib_price FLOAT64 DEFAULT 0.016;
DECLARE active_physical_gib_price FLOAT64 DEFAULT 0.04;
DECLARE long_term_physical_gib_price FLOAT64 DEFAULT 0.02;
SELECT
table_schema AS dataset_name,
ROUND(SUM(active_logical_bytes) / POW(1024, 3) * active_logical_gib_price
+ SUM(long_term_logical_bytes) / POW(1024, 3) * long_term_logical_gib_price, 2) AS logical_cost,
ROUND((SUM(active_physical_bytes) + SUM(fail_safe_physical_bytes)) / POW(1024, 3) * active_physical_gib_price
+ SUM(long_term_physical_bytes) / POW(1024, 3) * long_term_physical_gib_price, 2) AS physical_cost
FROM `region-us`.INFORMATION_SCHEMA.TABLE_STORAGE
WHERE table_type = 'BASE TABLE'
GROUP BY dataset_name
ORDER BY logical_cost DESC;Run the forecast over several weeks before switching, since a single snapshot taken after a large backfill or a cleanup can overstate or understate time travel bytes. When the numbers favor physical billing, the documented statement changes the model for one dataset:
ALTER SCHEMA mydataset SET OPTIONS(storage_billing_model = 'PHYSICAL');Updating dataset properties requires the bigquery.datasets.update permission, which the BigQuery Data Owner role includes. Because the change takes 24 hours to apply and blocks another change for 14 days, schedule it outside month-end reporting so the first bill under the new model is easy to compare. Teams that also want to keep query spend in check can pair the storage review with the habits described in writing SQL with Gemini in BigQuery and checking bytes scanned.
··········
COMMON MISREADINGS OF BIGQUERY PHYSICAL STORAGE BILLING.
A frequent assumption is that compressed storage is always cheaper because the byte count is smaller. The published rates contradict this for poorly compressed data: below about 1.74:1 compression, active physical bytes cost more than the same data billed logically, before any time travel or fail-safe charge is added.
A second misreading is that fail-safe can be turned off to save money. The time travel documentation states that the seven-day fail-safe period is not configurable, so the only lever under physical billing is the time travel window, between two and seven days. A third is that switching moves data back to active storage; the documentation states that long-term tables and partitions keep their long-term status after the change.
Finally, storage pricing applies only to data stored in BigQuery. Data read from external sources such as Cloud Storage, Bigtable or Google Drive is billed by those services, and long-term pricing does not apply to it. Extracts pulled into spreadsheets follow their own rules, covered in the guide to Connected Sheets limits and query costs.
··········
CHOOSING A BIGQUERY STORAGE BILLING MODEL FOR BI DATASETS.
The vendor confirms the four published rates, the 90-day rule for long-term storage, the free 10 GiB per month, the fact that time travel and fail-safe bytes are free under logical billing and charged at the active physical rate under physical billing, the 24-hour propagation and the 14-day wait between changes. Those facts make the decision a calculation that each team can run against its own metadata.
What depends on your environment is the compression ratio of each table, the share of data in long-term storage, how often tables are rewritten, the time travel window you can accept and whether legacy flat-rate commitments exist in the region. The rates quoted here are the ones listed for US locations, so the forecast query should use the rates the pricing page lists for the location where the dataset lives.
For the illustrative 10 TiB dataset, physical billing saved between about $41 and $113 a month with low churn and cost up to about $63 more with heavy daily rewrites. Datasets dominated by append-only, partitioned history are the strongest candidates for physical billing, while staging datasets rebuilt every day are often cheaper under logical billing.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org




