Databend

Editor’s note: This Databend review is based on my hands-on evaluation of Databend Cloud and the open-source Databend engine (the v1.2.9xx release line) in October 2026. I weighed pricing, setup, loading and pipelines, search and AI features, the open-source versus Enterprise split, security, and the rivals worth trialing instead.

Quick verdict: I recommend Databend for data teams on AWS that want Snowflake-style SQL workflows on an open-source engine, with per-second compute and $23 per TB-month storage. I also recommend it to teams that want vector and full-text search next to their warehouse tables. The trade-off: Databend Cloud runs only in four AWS regions you cannot change after signup, and Personal plan support has no SLA.

Databend gives you warehouses, stages, COPY INTO, streams, tasks and Time Travel on an Apache 2.0 core, so it is more than a cheap landing zone for batch SQL. Self-hosters get that engine free, but streams, masking and other governance features need an Enterprise license key.

In this review, I’ll take a closer look at Databend’s pricing, features, and open-source licensing, so you can see exactly whether it deserves a place on your shortlist next to Snowflake and ClickHouse. My recommendation favors teams comfortable with AWS-only hosting, not buyers who want a drop-in replacement for either rival.

Key Takeaways:

• Per-second compute starts at $1.00/hour for Personal XSmall and $1.50/hour for Business XSmall.
• Storage costs $23 per TB-month on Personal and Business.
• Cloud runs only in four AWS regions, fixed at signup: us-east-2, us-west-2, ap-southeast-1, and eu-central-1.
• The core is Apache 2.0, but self-hosted streams, masking, Python UDFs, and data sharing require an Elastic License 2.0 key. Cloud includes these features.
• Personal has a 24-hour Severity 1 response target, without a support SLA.

Pros and Cons

How Much Does Databend Cost?

Databend Cloud has three plans, and warehouses bill per second only while they are running. The 30-day free trial allows one user per organization, so treat it as a solo technical evaluation, not a team rollout.

• Personal ($1.00/hour for XSmall): For individual developers and small businesses that can live without multi-cluster scaling or a support SLA.
• Business ($1.50/hour for XSmall): For teams that need multi-cluster scaling, AWS PrivateLink, and a support SLA. Compute costs 1.5 times Personal at every warehouse size, and API requests cost twice as much.
• Dedicated (custom pricing): For larger companies that want a dedicated metadata store and compute pool, Prometheus metrics, and an SLA of up to 99.9% uptime.

Self-hosted Community is free under Apache 2.0, for commercial and non-commercial use; you pay only for the infrastructure you run it on. Enterprise is a license of one year or more with no public price, and its 15-day trial license is separate from the Cloud trial.

Plan comparison	Personal	Business	Dedicated
XSmall compute per hour	$1.00	$1.50	Custom
Storage per TB-month	$23	$23	Custom
API requests per 10,000	$1	$2	Custom
Multi-cluster scaling	No	Yes	Yes
Severity 1 support response	24 hours	8 hours	4 hours
Support SLA	No	Yes	Yes

Warehouse prices double at each size step, from $1.00 to $512 an hour on Personal. I recommend starting at XSmall or Small and moving up only when your queries run for several minutes.

Warehouse size	Personal per hour	Business per hour	Dataset sizing guidance
XSmall	$1.00	$1.50	Around 50 GB
Small	$2.00	$3.00	Around 200 GB
Medium	$4.00	$6.00	Around 1 TB
Large	$8.00	$12.00	Around 5 TB
XLarge	$16.00	$24.00	10 TB or more
2XLarge	$32.00	$48.00	Contact sales
3XLarge to 6XLarge	$64.00 to $512.00	$96.00 to $768.00	Contact sales

Is Databend Good Value for Money?

• Occasional queries stay cheap: An XSmall Business warehouse that runs one 20-minute query a day, with 100 GB stored, costs about $0.58 a day, or $17.40 a month. Idle minutes before auto-suspend and API requests come on top.
• Always-on workloads cost far more: Leave that warehouse running 24 hours a day with 1 TB stored and 50,000 API calls a day, and the bill reaches $46.77 a day, more than $1,400 over a 30-day month. Compute is $36 of the daily total and API requests $10, so I set auto-suspend explicitly; the default is five idle minutes, and you can adjust or disable it.
• Storage is priced at cost: $23 per TB-month follows Amazon S3 pricing and sits below ClickHouse Cloud’s $25.30. A $2.30 gap per TB favors storage-heavy workloads, but compute hours decide most bills.
• Connector hosting is billed separately: Data Integration adds $0.125 to $0.75 per hour, from 4XSmall to 2XSmall across Personal and Business. A CDC task left running around the clock on the smallest Personal size adds $90 a month, even if the warehouse barely runs.

Billing follows the calendar month, starting on the day you create the organization, and vouchers apply before paid charges, earliest-expiring first. Bills sit under Manage > Billing, and Databend Cloud is also sold through AWS Marketplace.

Reviewer’s Notes: I recommend starting with the trial, then choosing Personal for a single team in one region that can accept its support limits. Pay for Business when multi-cluster scaling, PrivateLink, or the 8-hour Severity 1 response becomes a requirement, not simply because the workload is production.

My Experience With Databend Cloud

Signing Up and Picking a Region

I registered at app.databend.com and created an organization, which made me its Admin. AWS Marketplace is the alternative route if you want the subscription on your AWS bill.

Organization setup asked for a name, a pricing plan, and a region on AWS, the only provider available:

• Ohio: us-east-2.
• Oregon: us-west-2.
• Singapore: ap-southeast-1.
• Frankfurt: eu-central-1.

The provider and region cannot change afterward, and organizations cannot be merged. Data is not shared across organizations by default; the separate data-sharing feature only works between tenants in the same region.

Trial organizations allow one user, so collaborative testing waits for a paid plan. Team invitations sit under Admin > Users & Roles, and an invited user cannot create a second organization.

Creating My First Warehouse

On the Warehouses page, I created an XSmall warehouse with auto-suspend on. The size menu runs from XSmall to XLarge, and 2XLarge and above require contacting Databend.

Each warehouse takes up to 10 tags (keys up to 128 characters, values up to 256), useful for marking environment and cost owner. A query against a suspended warehouse resumes it automatically, so there is no manual start step; for frequent queries, disabling auto-suspend keeps the cache warm.

Only account_admin can create or delete warehouses. Administrators can also run CREATE, ALTER, SUSPEND and RESUME WAREHOUSE in SQL, or bulk-suspend and resume from the console.

My main objection: warehouse-level role access is off by default. Turning it on takes a ticket through Support > Create New Ticket, which delays separating team access to compute.

Loading Data From S3

I created the destination table first, then ran COPY INTO with the S3 path, credentials, a file pattern and the format, rather than row-by-row INSERT.

COPY INTO books
FROM 's3://my-bucket/books/'
CONNECTION = (
  ACCESS_KEY_ID = '<your-key>'
  SECRET_ACCESS_KEY = '<your-secret>'
)
PATTERN = '.*[.]parquet'
FILE_FORMAT = (TYPE = 'PARQUET');

Databend supports schema detection and evolution, plus internal and external stages. Alongside Parquet, COPY INTO reads CSV, TSV, NDJSON, ORC and Avro, and unloads go back out as Parquet, CSV, TSV or NDJSON.

Running Queries in Worksheets

I opened Worksheets > New Worksheet (you can also start from a .sql file) and picked a database, role and warehouse. Cmd/Ctrl+Enter runs the statement under the cursor; Shift+Cmd/Ctrl+Enter runs all of them.

The result preview stops at 10,000 rows and 200 columns, and both caps are fixed. Cells show up to 3,000 characters by default, adjustable up to unlimited.

Exporting results to CSV needs the EXPORT permission, granted by the organization Owner. Shared worksheets let named members or the whole organization run queries without editing them.

For application access, BendSQL (the CLI) and drivers for Python, Go, Node.js, Java JDBC and Rust take over from the console grid. DBeaver, Tableau, Superset, Metabase and Grafana also connect.

Reviewer’s Notes: Get the region right on day one. A warehouse resizes with one ALTER WAREHOUSE statement, but an organization created in Frankfurt stays in Frankfurt, and since organizations cannot merge, a wrong pick means starting a new organization.

Loading and Pipelines: COPY INTO, Streams, Tasks and Data Integration

Databend Cloud’s Data Integration service runs no-code ingestion tasks, with hosting billed per second while a task runs:

• Amazon S3: One-time or continuous ingestion.
• MySQL and PostgreSQL: Snapshot, CDC only, or Snapshot plus CDC.
• Amazon SQS(S3), TiDB, and Kafka: Connectors marked Beta.

I favor Snapshot plus CDC for MySQL or PostgreSQL, since one task copies the existing tables and then keeps them current. The three Beta connectors stay off my production path until they lose that label, and their hosting fee lands on the pipeline bill even when the warehouse sits suspended.

Streams track table changes without duplicating storage. They default to append-only; set APPEND_ONLY=false for a standard stream that also tracks updates and deletes. Consuming a stream through a task, INSERT…SELECT or WITH CONSUME clears it.

Tasks run on an interval or CRON schedule, chain with AFTER, and can wait on a STREAM_STATUS guard before firing. SUSPEND_TASK_AFTER_NUM_FAILURES stops a task that keeps failing, and I set it before leaving any task unattended.

CREATE STREAM incoming_stream ON TABLE incoming;

CREATE TASK load_curated
  WAREHOUSE = 'etl_wh_small'
  SCHEDULE = 1 MINUTE
  WHEN STREAM_STATUS('incoming_stream') = TRUE
AS
  INSERT INTO curated
  SELECT * FROM incoming_stream;

The example assumes both tables and the warehouse already exist; in production I list columns explicitly instead of SELECT *.

Cloud’s Task Flow replaces the legacy Task List with a visual DAG editor, version history and stream-based triggers. For dependencies that are all SQL, it covers the job before you add a separate orchestrator.

Snowflake users will recognize streams and tasks; ClickHouse uses materialized views for comparable ingestion-driven transformations, with different change-tracking semantics.

For dbt, dbt-databend-cloud supports tables, views, incremental models and seeds, but not ephemeral models or snapshots. That blocks a migration if your existing project leans on either.

Other documented integration paths:

• Airbyte, DataX and Addax for data movement.
• Debezium and Flink CDC for database change pipelines.
• Kafka and Vector for event and log ingestion.

Search, Vectors and AI in the Same Engine

Vector and Full-Text Search

Databend stores embeddings in a VECTOR(n) column with a VECTOR INDEX using cosine distance, queried in the same SQL as your warehouse tables. When embeddings and analytical filters belong in one query, that removes the case for a separate vector database.

Text, location and nested data get their own index and type families in the same engine:

• Inverted indexes: MATCH() for relevance-ranked full-text search.
• NGram indexes: Speed up substring predicates such as LIKE ‘%term%’.
• Geospatial types and functions: Location-aware analysis.
• VARIANT, ARRAY, and MAP: Semi-structured values next to relational columns.

I like the fit for document retrieval with SQL filters: PDFs and images in S3 go through OCR, entity extraction and embeddings, then full-text or vector search inside Databend. Run a sample of your own scanned files through it before trusting the OCR output at scale.

ClickHouse and StarRocks also offer vector indexes, so vector support alone is not my reason to pick Databend. The bundle of search with Snowflake-style stages, streams and tasks is.

UDFs, MCP and the Cloud Agent

Databend runs SQL, Python and JavaScript UDFs, plus WebAssembly UDFs built from Rust modules on a stage (added in v1.2.936). Python UDFs need an Enterprise license when self-hosted, so a free Community deployment loses them.

External functions call HTTP services, including models on your own GPUs. The model runs on infrastructure you operate, so GPU capacity is your cost, not a Databend Cloud line item.

AI client connections:

• Hosted Cloud MCP: Browser OAuth sign-in for Claude, Cursor, and Codex.
• mcp-databend: A local MCP server for self-hosted deployments.
• Cloud Agent: A console chat assistant that runs SQL, draws charts and manages resources, marked Preview.

I keep the Preview-labeled Cloud Agent out of production requirements and build automation on MCP, BendSQL or the drivers.

Lakehouse Tables

Databend queries Iceberg tables through a catalog (support updated in v1.2.725) and handles Hive catalogs. A Docker Iceberg quick-start with Spark, a REST catalog, MinIO and TPC-H sample data lets you test catalog reads before pointing Databend at your own lake.

Open Source vs Enterprise: What the License Key Unlocks

Databend ships one binary: the core is Apache 2.0, while code in src/query/ee and src/meta/ee is Elastic License 2.0. Some Elastic-licensed features run without a key; the paid ones need an Enterprise license.

Databend Cloud includes every Enterprise feature, while a free Community download leaves this list locked:

• Pipeline and sharing: Streams, data sharing, and ATTACH TABLE.
• Governance: Masking policies, data lineage, and a trail of user activity.
• Custom processing: Python UDFs and virtual or computed columns.
• Protection: Storage encryption, BendSave backups, and Fail-Safe.
• Maintenance: Advanced vacuum operations for temporary files, dropped objects, and historical data.

Enterprise licenses run for one year or longer, with a 15-day trial, and non-commercial academic research can get long-term licenses. Install the key with SET GLOBAL enterprise_license='<key>', then check it with call admin$license_info().

ELv2 permits internal use and embedding in your own SaaS dashboards. It bars offering Databend as a hosted service with substantial access to its functionality and bars working around the license key, so review that boundary before building a managed database product on it.

The Rust project has 9,472 GitHub stars, 901 forks and about 261 contributors since October 2020. It shipped 30 releases between July 20 and October 21, 2026, including v1.2.954-nightly on October 21 and the v1.2.925-patch-15 tag on October 15.

The open issue tracker tempers that pace. October 2026 reports describe canceled queries leaking queue entries and canceled table-lock waits leaving locks that block writers, and an April report covers meta pods restarting under Kubernetes health checks.

Test query cancellation and pod recovery before any self-hosted production rollout. StarRocks and Apache Doris are the open-source engines to compare if you want governance features without a license key.

Security, Access Control and Compliance

Databend Cloud lists SOC 2 Type II and GDPR for all three editions, so the edition choice turns on the controls below:

• Access control: RBAC and discretionary access control, with masking and password policies.
• Network policies: ALLOWED_IP_LIST and BLOCKED_IP_LIST, with the block list checked first.
• User attachment: One network policy per user, although a policy can serve multiple users.
• Transport security: TLS 1.2.
• Private networking: AWS PrivateLink on Business and Dedicated.

Storage encryption under Databend Enterprise supports AES-256, customer-managed keys and HSM integration. Confirm which key-management option your edition gets before you sign, since encrypted storage and holding your own keys are different requirements.

Standard Time Travel retains up to one day; Business and Dedicated extend that to 90 days, and every edition adds seven days of Fail-Safe. That 90-day window is the strongest reason a recovery-sensitive team pays the Business premium.

Databend Cloud does not document an SSO setup, so ask sales which identity providers it supports before procurement. Plan the support ticket for warehouse-level role access into the same security rollout.

How Does Databend Compare to Competitors?

I recommend Databend for AWS-based, Snowflake-style SQL workflows with access to an open-source engine; these alternatives win when deployment flexibility, real-time analytics or very small workloads come first.

• Snowflake: My pick when you need AWS, Azure and Google Cloud instead of four AWS regions, or have no interest in self-hosting. Databend copies its warehouses, stages, streams and tasks on an open-source engine, but similar objects do not guarantee a painless migration, so test your own SQL.

• ClickHouse Cloud: I favor it for real-time operational analytics and its mature OLAP ecosystem. In us-east-1, Basic starts at $0.2181 per unit-hour and Scale at $0.2985, with storage at $25.30 per TB-month and a $300 trial credit. A unit-hour is not a Databend warehouse-hour, so compare workload bills, not headline rates.

• VeloDB: My pick when the data plane must live in your own cloud account. The Apache Doris-based service starts at $0.134 per vCPU-hour for Standard SaaS compute or $0.11 for BYOC, with a 30-day trial and $300 in credits. Separate storage and cache meters make the compute rate only part of the bill.

• StarRocks: I recommend shortlisting this open-source engine for self-hosted analytics without Databend’s Enterprise key boundary. The trade-off is owning the infrastructure instead of buying Databend Cloud’s managed warehouse.

• MotherDuck: My pick for small DuckDB-sized workloads that do not need a distributed warehouse. Lite starts free, while Business costs $250 per organization a month plus usage, and storage runs $0.043 per GB-month, roughly $44 per TB against Databend’s $23.

• BigQuery: I favor it when you never want to size a warehouse. On-demand queries cost $6.25 per TiB scanned, with the first TiB free each month. Databend ties compute spend to running time instead, so the choice turns on scan volume versus warehouse utilization.

How I Reviewed Databend

I evaluated Databend Cloud’s signup, warehouse, worksheet and S3 loading flows, and priced all three Cloud plans plus the self-hosted Community and Enterprise editions. My cost comparison separates compute, storage, API requests and connector hosting rather than stopping at the entry hourly rate.

I weighed streams, tasks and Data Integration, vector and full-text search, the AI interfaces, the Apache 2.0 versus Elastic License split, security and compliance, and support commitments. For self-hosting, I also went through the public GitHub repository, its release history and the open issue tracker.

I compared Databend with Snowflake, ClickHouse Cloud, VeloDB, StarRocks, MotherDuck and BigQuery on pricing, ease of setup, feature depth, openness and support.

Prices current as of October 2026.

Should You Run Your Warehouse on Databend?

Yes, for AWS-based teams that want Snowflake-style warehouses, pipelines and search without giving up access to an open-source core. XSmall compute at $1.00 to $1.50 per hour and storage at $23 per TB-month make Databend worth a workload trial, particularly when warehouses can suspend between jobs.

I recommend Personal for modest workloads with flexible support needs, and Business when multi-cluster scaling, PrivateLink, 90-day Time Travel or an 8-hour Severity 1 response is necessary. The gap between $17.40 a month for a mostly idle warehouse and $46.77 a day for an always-on one shows why auto-suspend and connector hosting matter as much as warehouse size.

Skip Databend Cloud if you need Azure or Google Cloud, a later region change, or fast contractual support on the entry plan. Compare StarRocks or Apache Doris if paid self-hosted governance features undermine the reason you want an open-source engine.

Start the Databend 30-day trial in the region you intend to keep, load a representative Parquet sample with COPY INTO, and set auto-suspend before leaving the warehouse running.

FAQs

How much does Databend Cloud cost?

Databend Cloud starts at $1.00/hour for Personal XSmall or $1.50/hour for Business XSmall, billed per second while running. Both charge $23 per TB-month for storage, plus $1 or $2 per 10,000 API requests and separate Data Integration hosting fees. Dedicated has custom pricing. Compare total running time, not just the starting hourly rate.

Is Databend open source?

Databend’s core engine is Apache 2.0, but Enterprise features use Elastic License 2.0. Self-hosted streams, masking, Python UDFs, and other paid capabilities require a license key; Databend Cloud includes all Enterprise features. Check the feature boundary before assuming a Cloud workflow transfers to a free Community deployment.

Does Databend have a free trial?

Databend Cloud offers a 30-day free trial limited to one user. Self-hosted Enterprise has a separate 15-day trial license, while Community is free to use under Apache 2.0. I recommend the Cloud trial for individual evaluation and a paid plan once you need team invitations and collaboration.

Does Databend Cloud run on Azure or Google Cloud?

No, Databend Cloud runs only on AWS, in Ohio, Oregon, Singapore, and Frankfurt. Azure and Google Cloud are planned but not available. The provider and region are chosen during organization setup and cannot change afterward, so pick the location of your intended production deployment, even for the trial.

Is Databend compatible with Snowflake?

Databend has similar SQL objects and workflows, not a guaranteed drop-in Snowflake replacement. Familiar capabilities include warehouses, stages, COPY INTO, streams, tasks, and Time Travel. Its dbt adapter does not support snapshots or ephemeral models. Test your actual SQL and dbt project before planning a migration.