ClickHouse Cloud vs Google BigQuery vs Snowflake: Real Analytics Warehouse Cost per 100 TB Scanned in 2026
The short answer: only BigQuery's on-demand model charges for bytes scanned, so "cost per 100 TB scanned" is a real price on exactly one of these three platforms. BigQuery on-demand bills $6.25 per TiB after a free first TiB each month: 100 TiB scanned costs $618.75, plus about $205 to store 10 TiB, or roughly $824 a month, regardless of how the queries are spread across the month. ClickHouse Cloud and Snowflake bill the compute that is running, not the data scanned, so the same 100 TiB scan can cost very different amounts depending on how many hours of compute are needed. Under an illustrative bursty pattern (a Medium-class warehouse billed 120 hours a month) the modeled bills are about $1,498 on Snowflake Enterprise and about $413 on ClickHouse Cloud Scale; under a sustained pattern (the same compute billed all 730 hours) they become about $8,818 and $1,870. The scan volume is identical in both patterns. Utilization, not volume, decides the bill.
Why this comparison is structurally difficult
BigQuery on-demand prices the work a query requests (logical bytes read from the columns it touches). Snowflake prices the time a virtual warehouse is running, whatever it is doing. ClickHouse Cloud prices the time a service is running at a given size, per minute. Converting the latter two into a dollar-per-terabyte number requires assuming how long compute runs, how big it is, and whether it is idle between queries; those assumptions are the answer. A fair comparison therefore needs two scenarios, one with low duty cycle and one sustained, and should treat any "$ per TB scanned" figure for Snowflake or ClickHouse as a derived average, not a price.
What each vendor actually bills
BigQuery (OFFICIAL, Google Cloud BigQuery pricing). On-demand: $6.25 per TiB of data processed, first 1 TiB per month free; a query's cost does not depend on its duration. Capacity (editions): billed per slot-hour, about $0.04 on Standard edition and $0.06 on Enterprise edition, with autoscaling and optional commitments; the default on-demand pool is 2,000 slots shared across a project. Storage: $0.02 per GiB-month for active logical storage (long-term storage, untouched for 90 days, is half that). Cached query results are free, and partitioning and clustering reduce bytes scanned, so scanned volume is a design variable, not a fixed property of the dataset.
Snowflake (OFFICIAL, Snowflake pricing; US AWS on-demand figures). Credits cost about $2 on Standard, $3 on Enterprise and $4 on Business Critical. A warehouse burns credits per hour by size: X-Small 1, Small 2, Medium 4, Large 8, doubling each step. Warehouses bill per second with a 60-second minimum on each resume, and auto-suspend after a configurable idle period. Storage is about $23 per compressed TB-month on a capacity contract and roughly $40 on-demand. Capacity contracts reduce credit prices by 20–40%. Snowflake's April 2026 AI Credits change affects Cortex agent features, not warehouse credits.
ClickHouse Cloud (OFFICIAL, clickhouse.com/pricing). Compute is metered per minute in compute units of 8 GiB RAM and 2 vCPU: $0.2181 per unit-hour on Basic, $0.2985 on Scale and $0.3903 on Enterprise (US East; other regions run higher, for example about $0.435 on Enterprise in eu-west-1). Storage is $25.30 per compressed TB-month on every tier, and backups bill at the same rate. Services scale down after roughly 15 minutes of inactivity, and a suspended service wakes more slowly than a Snowflake warehouse. Basic is capped in size and retains only one day of backups, which does not fit a moderate-concurrency production setup; Scale and Enterprise add autoscaling and multi-zone replicas.
Normalizing the workload
All ILLUSTRATIVE. 100 TiB of logical data scanned per month across about 10,000 queries (about 10 GiB per query); a 10 TiB logical dataset; moderate concurrency; one production environment. For the platforms that store compressed data, a 4:1 compression ratio gives 2.5 TB stored. Compute for Snowflake is a Medium warehouse (4 credits/hour); for ClickHouse it is 8 compute units (64 GiB RAM, 16 vCPU) on Scale or Enterprise, an assumption chosen to be of broadly comparable size, not a benchmark-derived equivalence. For BigQuery's capacity model, a 100-slot autoscaled reservation is assumed. Two utilization patterns: A, bursty/low duty cycle, 120 billed compute hours a month; B, sustained, all 730 hours. Cloud-services credits on Snowflake and egress on every platform are excluded.
Cost at the normalized workload
Formulas: BigQuery on-demand = (TiB scanned − 1) × $6.25 + GiB stored × $0.02; BigQuery capacity = slots × hours × $/slot-hour + storage; Snowflake = credits/hour × hours × $/credit + TB × $23; ClickHouse = units × $/unit-hour × hours + TB × $25.30 × 2 (data plus one backup-sized copy).
| Platform / model | Scenario A: bursty (120 h) | Scenario B: sustained (730 h) |
|---|---|---|
| BigQuery on-demand | $823.55 ($618.75 scan + $204.80 storage) | $823.55 |
| BigQuery capacity, Standard (100 slots) | $684.80 ($480 + $204.80) | $3,124.80 ($2,920 + $204.80) |
| Snowflake Enterprise, Medium ($3/credit) | $1,497.50 ($1,440 + $57.50) | $8,817.50 ($8,760 + $57.50) |
| Snowflake Standard, Medium ($2/credit) | $1,017.50 ($960 + $57.50) | $5,897.50 ($5,840 + $57.50) |
| ClickHouse Cloud Scale (8 units) | $413.06 ($286.56 + $126.50) | $1,869.74 ($1,743.24 + $126.50) |
| ClickHouse Cloud Enterprise (8 units) | $501.19 ($374.69 + $126.50) | $2,405.85 ($2,279.35 + $126.50) |
Dividing each bill by 100 TiB gives the effective cost per TiB scanned, using the full bill including storage: Snowflake Enterprise $14.98 in Scenario A and $88.18 in Scenario B, ClickHouse Scale $4.13 and $18.70, BigQuery on-demand about $8.24 including storage in both. The Snowflake and ClickHouse numbers move by roughly a factor of six with duty cycle, while BigQuery on-demand does not move at all.
Where on-demand beats capacity, and the reverse
Formula: on-demand scan cost = capacity cost when 99 TiB × $6.25 = slots × hours × $0.04. With 100 slots on Standard edition, that equals about 155 billed hours a month. Below roughly 155 hours of 100-slot activity, capacity is cheaper than on-demand; above it, on-demand is. In Scenario A (120 hours) capacity wins by about $139; in Scenario B (730 hours) on-demand wins by about $2,300. The result depends entirely on the slot count assumed: BigQuery capacity is only a better deal for a workload that scans a lot per slot-hour used.
The levers each platform exposes
On BigQuery the lever is bytes scanned: partitioning, clustering, column selection (avoiding SELECT *), materialized views and result caching can cut the 100 TiB without changing the workload's usefulness. On Snowflake the levers are warehouse size, auto-suspend timing and the 60-second resume minimum: a Medium warehouse resumed 10,000 times for 20-second queries can bill closer to 60 seconds each than 20, which is why bursty usage should be modeled by resumes, not query seconds. On ClickHouse the levers are replica count, replica size and idle timeout, and Scale's autoscaling will by default scale up to meet demand unless limits are set.
Sensitivity
- Duty cycle (billed compute hours). The dominant lever for Snowflake and ClickHouse: 120 versus 730 hours moves Snowflake Enterprise from about $1,498 to $8,818.
- Bytes scanned per query. The only lever on BigQuery on-demand: halving scanned bytes through partition pruning halves the scan charge.
- Compute size equivalence. Treating a Medium warehouse and 8 ClickHouse units as equivalent is an assumption; if ClickHouse needs half as many units, its numbers halve, and if Snowflake needs a Large, its numbers double.
- Snowflake edition and contract. $2, $3 or $4 per credit, before the 20–40% reduction a capacity contract can bring.
Budgeting traps
- Quoting Snowflake or ClickHouse as a dollar-per-terabyte price. It is a result of runtime and duty cycle, not a rate.
- Assuming BigQuery on-demand is cheap because the dataset is small. Cost follows bytes scanned, so a poorly partitioned table scanned repeatedly can exceed a warehouse bill.
- Ignoring the 60-second resume minimum on Snowflake. Many short queries spaced beyond the auto-suspend window each bill a full minute.
- Forgetting ClickHouse backups and Basic-tier limits. Backups bill at the storage rate, and Basic cannot scale or retain more than a day of backups.
- Using logical and compressed sizes interchangeably. BigQuery's scan and storage prices apply to logical bytes; Snowflake's and ClickHouse's storage prices apply to compressed bytes.
What to ask before you buy
Run the real query set on each platform and record bytes scanned (BigQuery), warehouse-seconds including resume overhead (Snowflake) and compute-unit minutes (ClickHouse) before converting anything to a per-terabyte figure. Ask Snowflake whether a capacity contract changes the credit price, and ask ClickHouse and Google for the exact regional rates for your deployment.