Databricks: cloud data, analytics and AI infrastructure

Databricks supplies a cloud platform for preparing, querying, governing and applying machine learning to enterprise data. Teams can work with common datasets through SQL, notebooks, scheduled pipelines and application interfaces. Its lakehouse architecture combines analytical processing with governed storage, while newer database and AI products extend the platform into operational applications.[1]

For banks, capital-markets firms and insurers, Databricks positions this foundation around investment research, trade and execution analytics, risk modelling, financial-crime analysis and customer operations. These are applications built from an institution’s data, models and controls; the published financial-services examples describe intended uses rather than a standard brokerage or execution service.[2]

Genie Agents were previously called Genie Spaces. Their curated data-question interface sits alongside Genie One, a broader business-user interface, and Genie Code for technical work. The distinction matters when specifying whether a team needs governed analytics, a general workspace assistant or development automation.[8]

Databricks is offered on AWS, Microsoft Azure and Google Cloud. Feature availability varies by cloud and region, so the presence of a regional workspace does not establish that every database, AI or networking option is available there. The architectural and operational examples below use the current AWS documentation where identified.[10]

Unity Catalog distinguishes managed tables and volumes, where it controls both governance and the storage lifecycle, from external assets, where it governs access while storage remains externally managed. That choice affects how an institution assigns data ownership, lifecycle administration and integration responsibilities.[6]

Databricks publishes product tariffs and offers negotiated usage commitments. Its Databricks Unit, or DBU, measures processing consumption; a DBU is not one universal server-hour or one fixed subscription price. The applicable product, cloud, region and usage pattern determine the charge. Storage and networking can add separate costs.[15]

The AWS networking schedule marks per-GB private connectivity as billing, while per-hour private endpoint fees are labelled as expected in the future. The latter should therefore be treated as a planned charge in this dated schedule, rather than assumed to be an existing universal fee.[17]

The AWS trial documentation offers credits valid for 14 days, through express signup or AWS Marketplace. A Marketplace subscription, or an account with a payment method, can convert to PAYG when credits are exhausted. Cancellation guidance requires stopping compute and, where applicable, removing card details before cancelling. Personal-email trials can have tighter scaling and feature limits.[18]

The billable-usage system table provides account-wide consumption records, normally arriving within 12 hours, with longer delays possible for new workspaces. It is useful for cost allocation and review, but that delivery delay means it should not be interpreted as an instantaneous spending meter.[21]

On AWS, the account controls identities, workspaces, metastore attachments and usage policies. Workspaces are the environments where teams run their workloads. Databricks manages the control plane; classic compute runs in the customer’s AWS account, while serverless compute runs in a Databricks-managed compute plane. The selected model changes who provisions and operates the underlying resources.[11]

A production design therefore needs an agreed cloud and workspace region, governed data locations, user and service-principal permissions, and a choice of classic or serverless resources. Databricks documents region-aligned compute planes and separate storage arrangements for classic and serverless workspaces; a generic “cloud deployment” description leaves these operational decisions unresolved.[11]

Data residency also depends on the service. Region-specific compute is distinct from Designated Services, including some AI capabilities, which process within a Databricks Geo comprising multiple regions. Cross-Geo processing can permit processing elsewhere and is enabled by default for certain workspaces outside the US and EU without a compliance security profile. Administrators need to inspect this setting for the actual workspace.[28]

Lakebase can synchronise Unity Catalog tables into Postgres for application reads and can connect through standard Postgres drivers. The reverse route that records Postgres changes as Delta tables is labelled Public Preview in the current documentation. Application integration and analytical replication therefore need to be scoped separately, including the release status of the chosen direction.[7]

Case submission requires a support contract and an authorised named contact. Customers normally submit through their workspace or manage cases in the Help Center. The documented email fallback is restricted to authorised contacts experiencing submission problems. An urgent case can be escalated to the on-duty manager; escalation retains the contract’s applicable initial-response commitment.[23]

The published direct-service MCSA, updated 20 February 2026, names Databricks, Inc. It supplies enterprise software services and allocates configuration, cloud-environment security, backups and data rights to the customer. Dedicated support is the level specified in the order; absent that, the agreement limits support to documentation. A customer’s brokerage permissions and conduct obligations remain separate from these cloud-service arrangements.[25]

The publicly linked DPA sample is version 3, dated 21 July 2023. It describes Databricks as a processor when handling customer personal data, with the customer acting as controller or processor as applicable. The customer remains responsible for lawful collection and necessary notices or consents, while processing is governed by documented instructions and the applicable agreement.[26]

The AWS compliance security profile is a charged add-on, providing additional monitoring, hardened compute and update controls. Databricks requires the relevant profile settings before processing data under the listed supported standards, including PCI-DSS. Customers remain responsible for their own compliance and external-system configuration. Sensitive values should not be placed in resource names or similar administrative fields, which can fall outside the compliance boundary.[27]

Databricks was founded in 2013 and is headquartered in San Francisco. Its commercial platform grew from work associated with Apache Spark and other open-source data projects. That open-source heritage coexists with a separately contracted managed platform, implementation choices and paid services.[29]

Operator

Databricks, Inc. for the published direct-service MCSA; Azure Databricks is supplied under Microsoft arrangements.

Service scope

Cloud data engineering, analytical processing, governance, transactional databases and AI infrastructure for institutional applications.

Customer types

Banks; Capital-markets firms; Asset managers; Insurers; Enterprise data and AI teams

Delivery

SaaS; API

Pricing publication

Public and quoted pricing

Headquarters

San Francisco, United States

Platforms

Databricks SQL; Lakeflow; Unity Catalog; Lakebase Postgres; Genie Agents; Genie One; Genie Code; Agent Bricks

Integrations

AWS hosting; Microsoft Azure hosting; Google Cloud hosting; Power BI and Tableau through SQL warehouses; JDBC/ODBC; SQL REST API and Python connector; OpenSharing; Kafka and cloud-message sources through Lakeflow

Pricing scope

Public usage tariffs and negotiated commitments; product, cloud, region, storage and networking affect the total.

Implementation

Select cloud, region, compute model, data locations and access controls; AWS classic and serverless compute have different operational boundaries.

Support

Purchased support tiers and authorised contacts; Help Center and documented case-submission routes. Coverage and initial-response targets depend on the plan.

Permissions and regulatory scope

Enterprise cloud software and data processing. Customer configuration, data rights and financial-service permissions remain their own responsibilities.

Provider metadata checked 21 Oct 2026Source 1Source 2Source 3Source 4Source 6Source 7Source 8Source 9Source 10Source 11Source 12Source 14Source 15Source 16Source 23Source 25Source 29Source 30Source 31Source 32Source 33Source 34Source 35Source 36Source 38

Free Edition is a separate, non-commercial, quota-limited environment without guaranteed reliability, support or an SLA. Its documented limits include serverless-only compute, one small SQL warehouse and five concurrent job tasks. Exceeding fair-use limits can suspend compute until the allowance resets while retaining data and settings. Moving to the full platform requires a new trial account, so a learning workspace should not be assumed to upgrade in place.[19]

Release labels affect operational use. Databricks classifies Private Preview and Beta platform features as unsuitable for production and without an SLA; Public Preview and generally available features have different support treatment. Workspaces using compliance security profiles support only specified preview features. This distinction applies to individual capabilities, even when the wider product family is established.[37]

For AWS PAYG accounts, cancelling the Databricks plan is permanent and removes access to workspaces, notebooks and data. Removing an AWS Marketplace subscription only removes that payment route; it does not cancel the Databricks plan. The billing guide describes workspace-content deletion within 30 days and directs customers to stop related compute and cloud resources. Contract accounts must use their account team’s process.[20]