Free browser tool · Browser-only processing

Data Warehouse Sizing Calculator

Enter your daily row volume, row size, and retention period to get compressed storage estimates and monthly cost comparisons across Snowflake, BigQuery, Redshift Serverless, and Azure Synapse. No account required, no GB knowledge needed.

  • ✓Row count → compressed TB conversion
  • ✓Side-by-side cost for 4 platforms
  • ✓Platform recommendation by workload
All calculations run in your browser — no data is sent to our servers.
🏗️

Your cost estimate will appear here

Enter your data characteristics on the left and click Estimate costs

Storage Estimate
—
Raw storage
—
Compressed
—
Incl. overhead
Best value for your workload profile
—
—
Monthly Cost Estimate (us-east-1 / us-central1, Q1 2024 on-demand rates)
Platform Storage Compute Total/month
Retention Cost Curve — Total Monthly Cost vs. Retention Period

Storage cost grows linearly with retention; compute stays flat. The gap widens as you keep data longer.

Pricing disclaimer: Estimates are based on approximate us-east-1 / us-central1 on-demand rates as of Q1 2024. Actual costs depend on your cloud region, edition tier, negotiated contract pricing, query complexity, and optimization choices. Verify current rates on each vendor's pricing page before making purchasing decisions: Snowflake, BigQuery, Redshift, Azure Synapse.

Data Warehouse Planning — 3-Tool Workflow

Step 1 · Design
Star Schema Designer
Open tool →
Step 2 · Plan
ETL Pipeline Mapper
Open tool →
Step 3 · Estimate
DW Sizing Calculator
← You are here
DATA ENGINEERING · CALCULATOR GUIDE

Cloud Data Warehouse Sizing: From Row Counts to Monthly Costs

Use the free calculator above to get a cross-platform cost estimate before you commit to a cloud data warehouse. This guide covers the sizing formula, how each platform prices storage and compute, what columnar compression actually does to your TB count, and ten specific moves that cut your monthly bill.

What Is Data Warehouse Sizing?

Data warehouse sizing is the process of estimating how much storage and compute your cloud data warehouse will require, based on the volume, structure, and query patterns of your data. A sizing estimate translates business data characteristics — like "50 million order rows per year" — into concrete infrastructure requirements: gigabytes or terabytes of compressed storage, and an estimated monthly cost for each platform you're evaluating.

Two components drive every cloud DW bill. Storage cost scales with how much data you keep and for how long — it's predictable and grows linearly. Compute cost scales with how often and how complex your queries are — it's the variable that tends to surprise people. This calculator estimates both.

Three inputs drive 90% of your storage footprint: row volume (how many rows enter your warehouse per day), row width (average bytes per row), and retention period (how many years you keep the data). A fourth factor cuts that raw number substantially: columnar compression. Cloud data warehouses compress structured data 3–10×, so a 357 TB raw dataset becomes roughly 72 TB after compression.

Before you build, a sizing estimate tells you two things. First, whether the monthly cost fits your budget — a 10 TB warehouse costs approximately ten times more than a 1 TB warehouse, and that ratio applies to each platform's storage line. Second, which platform suits your workload — BigQuery on-demand is cheapest for infrequent queries on small data; Snowflake or Redshift Serverless pull ahead for consistent high-volume workloads.

Once you have a cost estimate from this calculator, the natural next step is designing the schema. The Star Schema Designer builds out the dimensional model, and the ETL Pipeline Mapper maps source columns to destination tables — the full three-step workflow from design to estimate.

How to Estimate Your Data Volume

The core formula behind this calculator:

Compressed Storage Formula
Compressed TB = (Daily Rows × Row Size B × 365 × Retention Yrs × Table Factor) ÷ Compression Ratio ÷ 10¹²

Then multiply by 1.15 for metadata overhead (micro-partitions, indexes, temp storage).

Step 1: Estimate your daily row volume

Add up new rows generated per day across all source tables. Fact tables — orders, events, transactions, log entries — dominate this number. Dimension tables are mostly static and relatively small.

Data typeExample daily volume
E-commerce orders (10 items/order)100K orders × 10 = 1M rows/day
Web clickstream events10M–100M events/day at scale
IoT sensor readings (1 device/min)500 devices × 1,440 = 720K rows/day
CRM activity logs50K–500K rows/day
Financial transactions100K–10M rows/day

Step 2: Estimate average row width

Row width depends on how many columns you store and what type they are. Thin rows (~100 bytes) have 5–10 columns of mostly integers and short codes — IoT sensor readings, event flags. Medium rows (~500 bytes) have 15–30 columns of mixed text and numbers — typical CRM or ERP records. Wide rows (~2 KB) have 50+ columns with long text fields or nested JSON — web logs, document stores.

When in doubt, use 500 bytes as the default for structured relational data. Validate against a sample from your source system once you have access.

Step 3: Apply compression

Cloud data warehouses store data in columnar format, which achieves much higher compression ratios than row-based databases. Typical ranges: 3× for wide or sparse data, 5× for typical structured tables, 10× for integer-heavy or highly repetitive data like IoT sensor readings. Snowflake averages 5× compression in practice; BigQuery averages 7×; Redshift ra3 averages 4–6×.

Worked mini-example (IoT company): 500 IoT devices, each sending 1 reading per minute = 720,000 rows/day, thin rows at 150 bytes each, 5 years retention, standard 5× compression. Raw: 720K × 150B × 365 × 5 ÷ 10¹² = 197 TB raw → 39 TB compressed → 45 TB with 15% overhead. That 45 TB lands on the cost table.

Understanding Cloud Data Warehouse Pricing Models

Each platform has two cost components: storage (what you pay to keep data at rest) and compute (what you pay to query it). The two scale independently, and the ratio between them depends entirely on your workload.

Snowflake

Storage: Approximately $23/TB/month for on-demand storage (AWS us-east-1, Standard edition). Snowflake stores data as compressed micro-partitions on S3 — the price you pay is for compressed, stored GB.

Compute: Billed in credits. One credit equals one hour of runtime for an X-Small warehouse. Warehouse sizes consume credits proportionally: XS is 1 credit/hour (1–5 concurrent queries), S is 2 credits/hour (5–20 concurrent queries), M is 4 credits/hour (20–100 concurrent queries), L is 8 credits/hour (100+ concurrent queries). Standard edition rate on AWS us-east-1: approximately $3.00/credit. The critical feature is auto-suspend — the warehouse shuts down when idle, stopping credit consumption immediately.

BigQuery

Storage: $0.020/GB/month for active storage (data queried in the last 90 days). Data not modified or queried for 90 consecutive days automatically drops to long-term storage at $0.010/GB/month — 50% less, with no configuration required, reversed as soon as the data is queried again.

Compute: Two models. On-demand charges $5.00/TB of data scanned per query — cost-effective for light workloads, unpredictable for heavy analytical workloads. Flat-rate buys slots (100-slot increments, approximately $1,700/month for 100 slots); cost-effective when queries run more than 8–10 hours per day. The crossover: if your queries scan more than 340 TB/month, flat-rate is cheaper than on-demand.

Amazon Redshift Serverless

Storage: $0.024/GB/month for managed storage, automatically distributed across S3-backed Redshift managed storage.

Compute: $0.375/RPU-hour (Redshift Processing Units, us-east-1). RPU count scales automatically between 8 and 512 based on workload. Billed per second (minimum 60 seconds). Unlike Snowflake's explicit auto-suspend, Redshift Serverless scales to zero RPUs when idle.

Azure Synapse Analytics

Storage: $0.023/GB/month for Synapse managed tables — comparable to Redshift.

Compute: Billed per DWU-hour. DW100c (the smallest Synapse SQL Pool tier): $1.51/hour. Synapse Serverless uses a pay-per-query model at $5/TB scanned, similar to BigQuery on-demand.

PlatformStorage/TB/moCompute unitCompute rate
Snowflake (Standard)$23.00Credit$3.00/credit
BigQuery$20.00 active / $10.00 long-termTB scanned$5.00/TB
Redshift Serverless$24.58RPU-hour$0.375/RPU-hr
Azure Synapse$23.55DWU-hour$1.51/hr (DW100c)

These rates are approximate as of Q1 2024, AWS/GCP us-east-1 on-demand pricing. Negotiated enterprise contracts typically reduce costs 20–40%. Verify current rates on each vendor's pricing page before making purchasing decisions.

Columnar Compression — Why Your Data Shrinks 5–10× in a Cloud DW

Cloud data warehouses store data in columnar format rather than row format. In a row-based database, each row sits contiguously on disk: CustomerID, Name, Country, Tier — all together. In a columnar store, all CustomerID values sit together, all Name values together, all Country values together. That single architectural difference produces dramatically better compression ratios.

The reason: columns contain homogeneous, often repetitive data. A Country column in a 50-million-row customer table might have only 200 distinct values. Run-length encoding and dictionary encoding represent the entire 50 million rows in a few thousand entries instead of 50 million strings.

Three algorithms do most of the work:

  • Dictionary encoding replaces repeated strings ("United States" → integer 0) with a lookup table. Best for low-cardinality text columns like Country, Status, Region.
  • Run-length encoding (RLE) stores (value, count) pairs when the same value repeats consecutively. Best for sorted time-series or status columns.
  • Delta encoding stores differences between consecutive values rather than the values themselves. Best for timestamps, numeric sequences, and IDs.
Data typeTypical compression ratio
Integer columns (IDs, quantities)6–10×
Short strings, low cardinality (status, country)10–20×
Long text, high cardinality (names, descriptions)2–4×
Timestamps5–8×
Decimal / float3–6×
Mixed structured table4–7× (default to 5×)

A practical benchmark: a raw 100 TB Oracle OLTP database typically lands at 15–25 TB when migrated to a columnar cloud data warehouse — before any tiering or partitioning optimizations.

Storage vs. Compute — The Two Cloud DW Cost Drivers

Cloud data warehouse costs have two fundamentally different components, and they scale independently of each other.

Storage cost scales with how much data you keep and for how long. It's relatively predictable and grows linearly. Storing 10 TB costs approximately 10× more than storing 1 TB. Retention policy is the main lever: cutting retention from 5 years to 2 years reduces your storage bill by 60%.

Compute cost scales with how often and how complex your queries are. It's highly variable — a single runaway analytical query on a 50 TB table can cost more than a week of normal usage. Compute is where "surprise" cloud DW bills come from.

When storage dominates cost: low query frequency (dashboards running 2–3 times per day, nightly batch reports); large historical data archives; BigQuery long-term storage where data not queried in 90+ days drops to $0.01/GB.

When compute dominates cost: high-frequency dashboards with 100+ concurrent BI users refreshing live; large on-demand scans (BigQuery queries against unpartitioned, 50 TB+ tables); complex dbt model runs with many dependencies.

Rule of thumb: For typical business intelligence workloads — 100–500 BI users, daily refresh cycle — expect a 60%/40% compute-to-storage split. For archive-heavy data stores with light querying, the split flips toward 20%/80%.

Most platforms offer storage tiers that move cold data to cheaper storage automatically. BigQuery long-term storage, Redshift S3 Intelligent-Tiering, and Snowflake's storage pricing for data not in active virtual warehouses can cut storage costs 40–60% for historical data with no performance impact on hot data.

Choosing the Right Cloud Data Warehouse for Your Workload

No single platform suits every workload. The right choice depends on your data volume, query patterns, team skills, and existing cloud vendor relationships.

Workload profileBest fitReason
Small data (<1 TB), low query volumeBigQuery (on-demand)Pay-per-query; $0 compute cost when idle; near-zero minimum commitment
High concurrency (50+ users, real-time dashboards)SnowflakeMulti-cluster warehouses handle concurrency without queue waits; auto-scaling
Heavy ETL + analytics on the same platformRedshift ra3RA3 nodes decouple storage and compute; Redshift Spectrum queries S3 directly
Microsoft ecosystem (Azure, Power BI)Azure SynapseNative Power BI integration; Azure AD auth; cheapest compute for Windows shops
Petabyte scale (100 TB+), infrequent scansBigQuery (flat-rate)Separation of storage and compute at massive scale; slots purchased once cover all queries
Startup or unknown growth curveSnowflake or BigQueryBoth offer true serverless scaling; no upfront capacity planning required
dbt-centric modern data stackSnowflake or BigQueryFirst-class dbt support; large community; best CI/CD tooling

Three things to avoid optimizing for in isolation: lowest storage cost (BigQuery and Synapse are virtually identical per TB — storage will not decide the platform); lowest on-demand compute cost (BigQuery on-demand looks cheap until your analysts run a few 50 TB scans); and familiar SQL dialect (most analysts adapt within a few days).

Switching cloud data warehouse platforms later is expensive but possible. Snowpark, BigQuery Omni, and Redshift Federated Queries reduce lock-in. If you're multi-cloud or cloud-agnostic, prefer the platform your existing cloud vendor offers — consolidated billing and native IAM authentication are real operational benefits.

For more context on the cloud data warehouse landscape, the data analysis methods guide shows where a data warehouse fits in a broader analytical architecture. If you're evaluating whether a managed DWaaS solution makes sense over building your own warehouse, that decision affects both the platform choice and your total cost of ownership.

10 Ways to Reduce Your Cloud Data Warehouse Bill

  1. Define a data retention policy. Keeping every row forever is the most common cause of runaway storage costs. Set retention periods by table: facts that drive daily reporting need 2–3 years; audit logs need 7 years; staging tables need 7 days. Automate deletion or archival so the policy actually runs.
  2. Use storage tiering for cold data. BigQuery automatically moves data untouched for 90 days to long-term storage at $0.01/GB — half the active rate, zero configuration, reversed on the next query. Redshift S3 Intelligent-Tiering moves cold objects to cheaper storage classes automatically. These require no query changes.
  3. Partition and cluster your tables. Partitioning fact tables by date (and clustering by common filter columns like customer_id or region) reduces the data scanned per query by 70–90%. On BigQuery, this directly reduces your per-query bill. On Snowflake, micro-partition pruning reduces the warehouse time needed per query.
  4. Auto-suspend warehouses immediately (Snowflake). The default auto-suspend is 10 minutes. Set it to 60 seconds for development warehouses and 5 minutes for production. At $3/credit, a medium warehouse (4 credits/hour) left running overnight burns $288 for nothing.
  5. Use query result caching. Snowflake caches query results for 24 hours — rerunning the same dashboard query is free. BigQuery caches results for 24 hours too. Avoid SELECT * or dynamic timestamps in dashboard queries, which bypass the cache and force a full re-scan on every refresh.
  6. Materialize expensive views as tables. A BI dashboard querying a complex 20-join view recalculates it on every refresh. Materializing the view as a table once per hour costs one warehouse minute but saves 50+ query executions per day. In dbt, this is the difference between a view and a table materialization.
  7. Right-size your Snowflake warehouse. Run queries on XS or S warehouses; scale up only when concurrency or query time demands it. Most analytical queries finish faster on a larger warehouse only if they're actually CPU-bound — test before assuming bigger is better.
  8. Filter early on BigQuery on-demand. On BigQuery, bytes scanned are priced at the query layer, not the row layer. Using WHERE date_column >= DATE_SUB(CURRENT_DATE(), INTERVAL 90 DAY) on a date-partitioned table scans 90/730 ≈ 12% of a 2-year table — cutting your bill 88% for that query type.
  9. Switch dbt full-refresh models to incremental. Full-refresh dbt models re-process every row in your source tables every run. Incremental models process only new or updated rows. Switching a 1 TB+ fact table from full-refresh to incremental cuts model run cost more than 90%.
  10. Monitor and alert on credit consumption. Snowflake: use Resource Monitors to cap credit consumption and alert at 80% of budget. BigQuery: use Cloud Billing budgets with alert thresholds. Neither platform caps spending by default — a runaway query or forgotten warehouse eats real money before anyone notices.

Frequently Asked Questions

How accurate is this data warehouse sizing calculator?

The calculator gives directional estimates, not exact quotes. Storage estimates are typically accurate within ±20% for structured relational data. Compute estimates vary more widely depending on query complexity, caching behavior, and concurrent user behavior. Use these figures for budget planning and platform comparison, then validate with a proof-of-concept before committing to a contract.

What compression ratio should I use?

Standard (5×) is the right default for typical structured relational data — CRM records, transactional data, ERP extracts. Use Light (3×) for semi-structured or wide data with many text columns. Use Aggressive (10×) only for narrow integer-heavy tables such as IoT sensor readings, event flags, or numeric time-series. When in doubt, 5× produces a conservative estimate that's unlikely to surprise you on the low end.

Does this calculator include networking or egress costs?

No. Cloud data warehouse egress costs — data leaving the cloud region — are not included. Egress is typically $0.08–$0.09/GB on AWS and GCP, but most analytical workloads keep data within the same region, making egress costs negligible. For multi-region warehouses or large result set exports, add 5–15% to cover networking.

How does BigQuery pricing work differently from Snowflake?

Snowflake charges for the time your virtual warehouse is running, regardless of how much data it queries. BigQuery on-demand charges per byte scanned per query. For light workloads with small data volumes, BigQuery is typically cheaper. For high-concurrency dashboards scanning terabytes per day, Snowflake's always-on warehouse model — combined with auto-suspend — becomes more predictable and often cheaper.

What is a Snowflake credit?

A Snowflake credit is the unit of compute consumption. One credit equals one hour of runtime for an X-Small (XS) virtual warehouse. Larger warehouses consume credits proportionally faster: Small uses 2 credits/hour, Medium 4, Large 8. At Standard tier on AWS us-east-1, one credit costs approximately $3.00. Credits are consumed only while the warehouse is active — auto-suspend stops consumption immediately when the warehouse goes idle.

What is BigQuery's long-term storage pricing?

BigQuery tracks each table partition separately. Any partition that is not modified or queried for 90 consecutive days automatically transitions to long-term storage pricing at $0.010/GB/month — 50% less than active storage at $0.020/GB/month. This transition is automatic, requires no configuration, and reverses as soon as the data is modified or queried again.

How do I estimate daily row ingestion if I haven't launched yet?

Estimate from the business operations your warehouse will track. An e-commerce company processing 10,000 orders per day with an average of 8 items per order generates 80,000 order line item rows per day — plus customer activity events, product views, and returns. Plan for 500K–2M rows per day at that order volume. Add a 50% buffer for year-one growth.

Should I include dimension table rows in my daily ingestion count?

Focus your daily ingestion count on fact table rows — these grow continuously and dominate storage over time. Dimension tables (customers, products, employees) are much smaller and mostly static. If your largest dimension tables exceed 50 million rows and update frequently, add 10% to your ingestion estimate to account for them.

What's the difference between Redshift Serverless and Redshift ra3?

Redshift Serverless automatically allocates Redshift Processing Units on demand, bills per second, and scales to zero when idle — the right choice for variable or unpredictable workloads. Redshift ra3 uses fixed provisioned node counts with separate managed S3 storage — better for consistent high-throughput workloads where reserved pricing (1 or 3 year) reduces cost by 40–60%.

Can I use this calculator for on-premises data warehouse sizing?

The storage volume estimates — raw GB and compressed GB — apply to on-premises sizing. The cost outputs apply only to the four cloud platforms listed. On-premises sizing involves hardware acquisition, rack space, power, cooling, and staffing costs that require a separate total cost of ownership model.

When this tool is useful

From row counts to a defensible budget number.

Every competitor calculator — Snowflake's, Google's, AWS's, Azure's — requires knowing your GB requirements before it can produce a cost estimate. This calculator starts from business data characteristics you already know: how many rows your systems generate, how wide those rows are, and how long you need to keep them. It converts that to compressed storage and a monthly cost across all four major cloud platforms, with a recommendation based on your specific query volume and concurrency.

Comparing cloud warehouse platformsFirst warehouse budgetsRetention policy decisionsMigration cost planning
FAQ

Data Warehouse Sizing Calculator questions

Common questions about how the calculator works.

How accurate are the estimates?+

They are directional planning figures, not quotes. Vendor prices change, so check each platform's pricing page before you commit.

Which compression ratio should I use?+

Standard (5×) suits typical relational data. Use 3× for wide text-heavy tables and 10× only for narrow, repetitive numeric data.

Are egress or networking costs included?+

No. Data transfer out of a cloud region is not part of the estimate.

Is my data sent to a server?+

No. The calculation runs in your browser.