If your work is moving towards data engineering and machine learning on open formats, look at Databricks. If you want a serverless warehouse with per-query billing on Google Cloud, BigQuery; on AWS with provisioned or serverless options, Amazon Redshift; and if your organisation already runs on Power BI and Azure, Microsoft Fabric. For low-latency, user-facing analytics, ClickHouse and Firebolt are built for that job. If the goal is to keep data in your own object storage and query it in place, Starburst (Trino) and Dremio are query engines rather than warehouses. For data that fits on one machine, DuckDB (and MotherDuck as its managed service) may remove the need for a cloud warehouse altogether.
Snowflake is a cloud data warehouse that separates storage from compute. Compute runs in virtual warehouses that consume credits while they are running; Snowflake documents per-second billing after a 60-second minimum each time a warehouse starts or resumes, with each larger warehouse size using roughly double the credits of the size below. Storage is billed on the average compressed volume per month. Credit prices depend on the edition (Standard, Enterprise, Business Critical or Virtual Private Snowflake), the cloud and the region, and Snowflake accounts run on AWS, Microsoft Azure or Google Cloud.
That model works well for many teams. Those who look elsewhere usually give one of these reasons; the first and last are based on Snowflake's own documentation, the others are our editorial reading rather than survey data.
- Credit costs are hard to predict. Spend depends on warehouse size, how long warehouses stay running, auto-suspend settings and serverless features, which are billed separately in compute-hours. Teams with many always-on dashboards or ad hoc users often find it hard to forecast.
- Lock-in to one platform. Data in standard Snowflake tables lives in Snowflake-managed storage and is queried through Snowflake. Moving away means exporting data and rewriting SQL, stored procedures and Snowpark code (editorial).
- Low-latency and real-time analytics. Customer-facing dashboards with many concurrent, sub-second queries are a different workload from batch reporting; specialised engines such as ClickHouse and Firebolt target it directly (editorial).
- Open formats. Snowflake supports Apache Iceberg tables, but it documents that when Snowflake is the Iceberg catalog, third-party engines cannot append, delete or modify the data. Teams that want several engines writing to one copy of the data look at lakehouse and query-engine options.
For head-to-head pages, see Snowflake vs Databricks, Snowflake vs BigQuery, Snowflake vs Redshift and ClickHouse vs Snowflake. If you are not yet sure you need a warehouse at all, start with data warehouse vs database.
Quick picks
Spark, SQL warehouses and ML on Delta Lake and Iceberg tables, on AWS, Azure and Google Cloud.
No warehouses to size: on-demand billing per TiB scanned or capacity editions with autoscaling slots.
One capacity covering warehouse, Spark, pipelines and Power BI, with OneLake storing tables as Delta or Iceberg.
Apache 2.0 open source engine with a managed service on AWS, Google Cloud and Azure.
Trino-based federated SQL over object storage and databases, managed or self-managed.
MIT-licensed in-process engine, with an optional managed service that has a free plan.
How we chose
We considered Databricks, Google BigQuery, Amazon Redshift, Microsoft Fabric, ClickHouse Cloud, Firebolt, Starburst, Dremio, DuckDB with MotherDuck, and Azure Synapse Analytics. To be included, a product had to be sold or actively developed in October 2026, handle analytical SQL over large data sets, and publish its pricing model and supported clouds on an official page. Azure Synapse was left out as a separate entry: Microsoft's documentation now directs dedicated SQL pool customers towards Fabric Data Warehouse with a migration assistant, so Fabric is the Microsoft option listed here (see Snowflake vs Azure Synapse for the existing product).
For each product we recorded the billing unit, the clouds it runs on, its support for open table formats (Apache Iceberg, Delta Lake), its free tier or trial, and its main limitation. Exact prices are not quoted on this page because they vary by edition, cloud and region; check the linked pricing pages and the vendors' calculators. None of this comes from our own use of the platforms.
The order is not a ranking. Entries start with full warehouse and lakehouse platforms, then specialised real-time engines, then query engines over your own storage, then single-machine options.
At a glance
| Platform | Pricing model | Clouds | Open formats | Best for | Main trade-off |
|---|---|---|---|---|---|
| Databricks | DBUs per second, plus cloud VM costs on classic compute | AWS, Azure, Google Cloud | Delta Lake; managed Iceberg tables | Data engineering, ML and SQL on a lakehouse | Two bills on classic compute; more to configure |
| Google BigQuery | Per TiB scanned (on-demand) or slot capacity editions | Google Cloud; Omni for some AWS and Azure regions | Iceberg managed tables in your Cloud Storage | Serverless analytics on Google Cloud | On-demand costs follow bytes scanned |
| Amazon Redshift | Node-hours (RG, RA3) or Serverless RPU-hours | AWS only | Reads Iceberg via AWS Glue Data Catalog | Teams standardised on AWS | AWS only; more tuning than serverless rivals |
| Microsoft Fabric | Capacity units (F SKUs), OneLake storage separate | Microsoft-hosted SaaS on Azure | Delta Parquet and Iceberg in OneLake | Power BI and Microsoft 365 shops | Capacity sizing; Power BI licences below F64 |
| ClickHouse Cloud | Compute units per minute plus storage | AWS, Google Cloud, Azure; BYOC | Reads and writes Iceberg (with limits) | Real-time and user-facing analytics | Different SQL dialect and data modelling |
| Firebolt | Per-second engine billing, storage pass-through | AWS and Google Cloud (GA), Azure (preview) | Reads Parquet and Iceberg | Low-latency, high-concurrency queries | Smaller ecosystem; open source edition in preview |
| Starburst | Credits (Galaxy) or licence (Enterprise) | Galaxy: AWS, Azure, Google Cloud; Enterprise: anywhere | Iceberg, Delta Lake, Hive, Hudi | Federated SQL over your own storage | A query engine, not a storage platform |
| Dremio | Dremio Compute Units (Cloud); Enterprise by quote | Cloud: AWS; Enterprise: AWS, Azure, Google Cloud, on premises | Iceberg, Apache Polaris catalog | Iceberg lakehouse with SQL acceleration | Dremio Cloud is AWS only for now |
| DuckDB and MotherDuck | DuckDB free; MotherDuck free plan then usage-based | Any machine; MotherDuck on AWS | Parquet, Iceberg (read and write via REST catalog), DuckLake | Data that fits on one machine | Not a multi-user enterprise warehouse |
Databricks
Databricks is a lakehouse platform: data is stored as Delta Lake or Apache Iceberg tables in cloud object storage and governed by Unity Catalog, and the same tables are used by Spark notebooks, jobs, pipelines and Databricks SQL warehouses. Databricks documents that managed Iceberg tables in Unity Catalog can be read and written by external engines through the Iceberg REST Catalog API, and that UniForm lets Iceberg clients read Delta tables. That is the main contrast with Snowflake for teams that want more than one engine on the same copy of the data.
Billing is in Databricks Units (DBUs), pay as you go per second, with committed-use discounts. On classic compute in your own cloud account, Databricks states that your cloud provider also bills you for the virtual machines; serverless compute runs without provisioning resources in your account. Databricks Free Edition, the successor to Community Edition, is for personal, non-commercial use. See Snowflake vs Databricks.
- On classic compute you pay Databricks for DBUs and your cloud provider for VMs, storage and networking, which makes cost tracking harder.
- More concepts to learn (clusters, jobs, notebooks, Unity Catalog) than a SQL-only warehouse.
- Free Edition does not allow commercial use, so evaluation for work needs a trial.
Google BigQuery
BigQuery is Google Cloud's serverless warehouse. There are no warehouses to start or size: on-demand compute is billed per TiB scanned, with the first 1 TiB of queries and 10 GiB of storage each month free, as listed on Google's pricing page on 7 October 2026. Alternatively, capacity pricing uses slots through the Standard, Enterprise and Enterprise Plus editions, with autoscaling and optional 1-year or 3-year commitments on Enterprise and Enterprise Plus. Storage can be billed on logical (uncompressed) or physical (compressed) bytes.
For open formats, BigQuery offers Apache Iceberg managed tables that store data in your own Cloud Storage bucket, which Google documents as readable by engines such as Spark. BigQuery Omni can query data in Amazon S3 (six AWS regions) and Azure Blob Storage (one Azure region). See Snowflake vs BigQuery.
- On-demand cost follows bytes scanned, so unpartitioned tables and SELECT * queries can be expensive.
- Primarily a Google Cloud service; Omni covers only a few AWS and Azure regions.
- The Standard edition lacks features such as BigQuery ML and column- and row-level security.
Amazon Redshift
Amazon Redshift offers provisioned clusters and Redshift Serverless. AWS now recommends RG nodes (Graviton-based, with an integrated data lake query engine) or RA3 nodes; both use Redshift managed storage, so compute and storage are billed separately, and provisioned clusters can be paused. Redshift Serverless bills compute in RPU-hours per second with a 60-second minimum, and AWS offers a free trial credit for new Serverless users.
Redshift can query Apache Iceberg tables catalogued in the AWS Glue Data Catalog and join them with local tables; on RG and Serverless these queries run on the warehouse's own compute, while RA3 uses Redshift Spectrum, billed per data scanned. It suits teams whose data, identity and tooling are already in AWS. See Snowflake vs Redshift.
- AWS only, so it does not reduce cloud lock-in.
- AWS documents that Python UDFs are no longer supported after 30 June 2026, which affects older Redshift code.
- Provisioned clusters need node type and size choices; Iceberg time travel queries are not supported.
Microsoft Fabric
Microsoft Fabric is a SaaS analytics platform that combines Data Factory pipelines, Spark data engineering, a T-SQL data warehouse, real-time analytics and Power BI. Compute is bought as capacity: F SKUs from F2 to F8192 capacity units, billed per second with a one-minute minimum on pay-as-you-go or discounted through reservations, and an F capacity can be paused. OneLake storage is billed separately.
OneLake stores tables in Delta Parquet or Iceberg format, and Microsoft documents that Delta tables can be read by Iceberg readers such as Snowflake through metadata virtualisation. Shortcuts reference data in ADLS, Amazon S3 and other locations without copying. For Synapse users, Microsoft provides a Fabric Migration Assistant for dedicated SQL pools. See Snowflake vs Azure Synapse and Databricks vs Microsoft Fabric.
- Every workload shares one capacity, so a heavy job can throttle others until you resize or pause.
- On F SKUs smaller than F64, users viewing Power BI content need Pro or Premium Per User licences.
- Runs only as a Microsoft service; it is not available on AWS or Google Cloud infrastructure.
ClickHouse Cloud
ClickHouse is a column-oriented analytical database under the Apache 2.0 licence, built for fast aggregation over large event and log tables. ClickHouse Cloud is the managed service, with Basic, Scale and Enterprise tiers on AWS, Google Cloud and Azure; compute is metered per minute in compute units and storage is billed separately, and a bring-your-own-cloud option is quoted on request. The trial runs for 30 days with credits and needs no card.
ClickHouse can read and write Apache Iceberg tables, with documented limits on some column types and on updates and deletes in format-version 3 tables. Teams typically move dashboards that need many concurrent, low-latency queries from Snowflake to ClickHouse and keep batch reporting elsewhere (editorial). See ClickHouse vs Snowflake.
- Its SQL dialect, table engines and sort keys need different modelling from Snowflake, so migration is more than a lift and shift.
- Frequent single-row updates and deletes are not its strength; it is designed for append-heavy analytics.
- Iceberg write support has documented type and DML restrictions.
Firebolt
Firebolt is an analytical database aimed at data-intensive applications that need many fast concurrent queries. Its documentation lists the managed service as generally available on AWS and Google Cloud and in preview on Azure, and describes a single binary that runs on a laptop, a VM or Kubernetes. Pricing is per-second engine billing with auto-stop and resume, plus object storage charged through at cost; new users receive trial credits.
It reads open formats (Parquet and Iceberg) and offers managed, bring-your-own-cloud and self-hosted deployment. Firebolt OSS, the self-hosted open source edition, was still in preview when checked.
- The open source edition is in preview, so self-hosting is not yet a GA option.
- Smaller ecosystem of connectors, tools and practitioners than Snowflake.
- Azure support is in preview.
Starburst
Starburst sells two products built on Trino, the open source distributed SQL engine governed by the Trino Software Foundation under the Apache 2.0 licence: Starburst Galaxy, a fully managed service, and Starburst Enterprise, which you deploy yourself. Galaxy bills compute in credits, with a free tier of up to three clusters and paid Pro, Enterprise and Mission-Critical tiers, plus a 30-day trial.
Starburst does not own your storage: it queries Iceberg, Delta Lake, Hive and Hudi tables in object storage and can join them with relational databases. That suits teams who want to stop loading everything into one warehouse. For the engine background, see Trino vs Presto.
- It is a query engine, so storage, table maintenance and catalog choices remain your responsibility.
- Federated queries across sources depend on the speed of those sources.
- Credit prices vary by tier, cloud and region.
Dremio
Dremio is a lakehouse query platform centred on Apache Iceberg and the Apache Polaris open catalog. Dremio Cloud is fully managed and billed in Dremio Compute Units, with a 30-day trial and no card needed; Dremio Enterprise is self-managed on Kubernetes, on premises or in the cloud; and a free Community Edition runs on your own machine or server.
It suits teams who want Iceberg tables in their own storage, queried by Dremio and by other engines, rather than data held in a proprietary warehouse format.
- Dremio Cloud runs on AWS only; Dremio lists Azure as coming soon.
- Enterprise pricing is by quote.
- You manage more of the lake (storage layout, catalog, table maintenance) than in Snowflake.
DuckDB and MotherDuck
DuckDB is an MIT-licensed analytical database that runs inside your process, with no server to manage. It reads Parquet and CSV directly and its Iceberg extension can read Iceberg tables and write to tables managed by an Iceberg REST catalog. The current stable release is 1.5.6, and DuckDB 2.0.0 is scheduled for 21 October 2026, so check the release notes before relying on version-specific behaviour.
MotherDuck is a managed service built on DuckDB, running on AWS in six regions. Its Lite plan is free (10 GB storage and 10 hours of Pulse compute a month, up to 3 users, as listed on 7 October 2026), Business adds usage-based compute and managed DuckLake, and Enterprise is custom. In our view, teams whose data is in the gigabytes rather than terabytes should check this option before choosing another cloud warehouse. See DuckDB vs Snowflake.
- DuckDB is single-node and in-process, so it is not a shared, multi-user warehouse by itself.
- MotherDuck runs only on AWS.
- A major DuckDB release (2.0) is imminent, so tooling and extensions may need updating.
When to stay with Snowflake
Stay with Snowflake if your main problem is cost and you have not yet tuned it: auto-suspend, right-sized warehouses, resource monitors and moving rarely queried data to Iceberg tables in your own storage often change the bill without a migration. Snowflake also runs on AWS, Azure and Google Cloud, which BigQuery, Redshift and Fabric do not, and in our view its data sharing and governance features are hard to replace once other teams depend on them.
Switching makes most sense when the workload itself has changed: heavy data engineering and ML (Databricks), a commitment to one cloud's ecosystem (BigQuery, Redshift, Fabric), sub-second customer-facing queries (ClickHouse, Firebolt), several engines on one copy of data (Starburst, Dremio), or data small enough for one machine (DuckDB). Expect to rewrite SQL dialect differences, stored procedures and any Snowpark code, and run both platforms in parallel before cutting over.
Frequently asked questions
What is the cheapest alternative to Snowflake?
It depends on data size and query patterns, not on list prices. For data that fits on one machine, DuckDB is free and MotherDuck has a free plan. For large data, compare on-demand BigQuery (billed per TiB scanned), Redshift Serverless (RPU-hours) and ClickHouse Cloud (compute units) using each vendor's calculator with your own workload.
Is Databricks a replacement for Snowflake?
For many warehouse workloads, yes: Databricks SQL warehouses run SQL over Delta and Iceberg tables. Databricks is broader (Spark, pipelines, ML) and has more to configure. See Snowflake vs Databricks.
Is there an open source alternative to Snowflake?
There is no single open source equivalent, but ClickHouse (Apache 2.0), DuckDB (MIT) and Trino (Apache 2.0, the engine behind Starburst) are open source engines, and Apache Iceberg is an open table format that several of them can read and write.
Does Snowflake support Apache Iceberg?
Yes. Snowflake can manage Iceberg tables as the catalog or use external catalogs such as AWS Glue or Snowflake Open Catalog. Snowflake documents that third-party engines cannot write to Iceberg tables when Snowflake is the catalog, while tables in an external catalog can be written by both.
What replaced Azure Synapse Analytics?
Microsoft positions Microsoft Fabric as the destination for Synapse dedicated SQL pool workloads and provides a Fabric Migration Assistant. See Snowflake vs Azure Synapse.
Do I need a data warehouse at all?
Not always. For modest data volumes, a PostgreSQL or SQL Server read replica, or DuckDB, may be enough. See data warehouse vs database, Snowflake vs PostgreSQL and Snowflake vs SQL Server.
Sources
- Snowflake pricing options
- Snowflake: understanding compute cost
- Snowflake: supported cloud platforms
- Snowflake: Apache Iceberg tables
- Databricks pricing
- Databricks: Iceberg in Unity Catalog
- Databricks Free Edition
- BigQuery pricing
- BigQuery editions
- BigQuery Omni
- Amazon Redshift pricing
- Amazon Redshift provisioned clusters and node types
- Amazon Redshift: Apache Iceberg tables
- Microsoft Fabric pricing
- Microsoft Fabric licences and capacity
- OneLake overview
- Migrating Synapse dedicated SQL pools to Fabric
- ClickHouse Cloud pricing
- ClickHouse Iceberg table engine
- Firebolt pricing
- Firebolt documentation
- Starburst pricing
- Trino Software Foundation
- Dremio pricing
- DuckDB Iceberg extension
- MotherDuck pricing
Checked October 2026.
How we research these guides: our editorial method.