Snowflake: institutional data, analytics and AI infrastructure

Snowflake supplies a managed cloud platform for storing, processing and analysing institutional data. Its architecture separates storage, compute and cloud services, allowing workloads to use independently configured compute against governed information. Snowflake operates the underlying service and maintenance. The core platform is not software that a brokerage installs on its own premises or private cloud.[1]

For asset and wealth managers, Snowflake positions the platform around market, portfolio and client information, quantitative research, risk analysis and reporting. Firms can bring proprietary and externally sourced datasets into these workflows. These are documented technology use cases; the investment process and decisions remain with the institution using the data.[2]

An account is provisioned on a selected AWS, Microsoft Azure or Google Cloud region. That choice hosts the account’s storage, compute and cloud-services layers and affects unit costs. Data can be loaded from, or unloaded to, supported cloud object stores irrespective of the account’s host. The data-source cloud and the Snowflake account cloud therefore need not be the same.[4]

Same-region direct sharing exposes selected objects as read-only data without creating a storage copy in the consumer’s account. The consumer pays for the compute used to query it. The provider controls which accounts can access the share, can change available objects and can revoke access. This access model is useful for distributing governed information while retaining source-side control.[9]

Sharing across a cloud or region boundary uses a different route. The documented replication workflow creates source and target replication groups, refreshes the target data and shares from the destination account. Replicating a purpose-built subset database can reduce the scope. Cross-region delivery through this route therefore involves a replicated copy and its operating costs; the same-region no-copy description does not apply globally.[10]

Snowflake’s managed MCP server exposes Cortex Analyst, Cortex Search, Cortex Agents and configured custom or SQL tools to compatible clients. Snowflake operates the server infrastructure. Authentication uses Snowflake OAuth by default, with External OAuth available, and roles determine server and tool discovery and invocation. This connects an AI client to approved capabilities without making every account object available to that client.[15]

Horizon’s open catalogue interfaces support Iceberg access through a REST API, including engines such as Spark, Flink and Trino. The documentation also describes integration with external catalogues such as AWS Glue and Azure OneLake. These interfaces connect data estates, while the chosen catalogue, table type and access controls determine the actual read/write and governance behaviour.[14]

Snowflake combines usage-based compute with storage and other service charges. Customers can use on-demand billing or purchase capacity in advance. Storage is measured by the average compressed amount retained during the month. Edition, cloud and region affect the available features and commercial configuration; a single seat price does not describe the platform’s production cost.[5]

The Service Consumption Table effective 15 October 2026 distinguishes Platform Credits from AI Credits. Platform-credit discounts do not automatically apply to AI credits. AI billing also depends on the model’s token, compute-time or other specified usage units.[7]

Cloud-services processing and serverless feature compute are separately metered. Data-transfer fees depend on the cloud, region and destination, including transfer between certain compute components. These charges sit alongside warehouse consumption.[7]

For a conventional warehouse, size, active clusters and running time drive compute use. Suspending idle resources avoids continued warehouse consumption, while starting or resuming introduces the applicable minimum period. Sizing should follow the workload and concurrency requirement, rather than the number of stored rows alone.[6]

Warehouse resource monitors can notify at credit thresholds, suspend after running queries complete or suspend immediately. Auto-suspend/resume and statement timeouts provide further controls. The monitor mechanism covers virtual warehouses, not every cloud-services or serverless charge. Controlling an analytical warehouse therefore does not cap the entire account’s expenditure.[8]

The public evaluation offer provides $400 of credits over 30 days.[2]

Production fees and payment terms are set by the applicable order form. The Terms of Service make obligations non-cancellable and fees non-refundable except where the agreement or law provides otherwise; taxes are additional.[21]

Access design begins with the organisation, accounts, databases, schemas and objects. Roles and privileges determine who can use warehouses, inspect data and administer resources. In a managed-access schema, grant authority sits with the schema owner or a role holding MANAGE GRANTS, rather than each object’s owner. That distinction affects how platform teams delegate control to research, operations and application teams.[17]

Ingestion design selects the bulk-file, continuous-file or streaming-row path before downstream modelling. File loads can apply limited transformations during ingestion, while dynamic tables, streams and tasks support subsequent processing. Warehouse sizing for bulk loading and Snowflake-managed continuous ingestion represent different operational and billing choices; neither should be inferred from a generic claim that the platform is serverless.[16]

The standard Terms of Service provide a data-retrieval right when the customer gives written notice: access for up to 30 calendar days after expiry or termination, solely to retrieve Customer Data. The agreement and order forms continue during that period. After the later of termination and the retrieval period, Snowflake promptly deletes remaining Customer Data. The contractual window therefore needs to align with the technical export process.[21]

The published standard support offering is included without an additional support fee and permits unlimited submitters. Existing users open cases through the service interface or the Snowflake Community portal. The legal policy makes the purchased support level an order-form term and allows more limited support for evaluation use.[23][24]

These are initial-response targets, not resolution deadlines. Availability credits follow the policy’s monthly-availability and error-rate matrix, with separate feature-level rules, including for Postgres. They are service credits rather than cash refunds. The policy also contains a time-limited claim process and exclusions for trial and preview use.[24]

This profile covers data infrastructure and analytical software. Financial-services workflows can combine market data, client information and investment models, but the described platform does not itself establish a brokerage account or an investment mandate. The institution supplies the business process and the relevant permissions for its activities.[2]

The customer and its licensors retain ownership of Customer Data. Snowflake receives the processing rights needed to deliver and troubleshoot the service or meet legal requirements. Customers must have sufficient rights to provide the data and remain responsible for its legality and accuracy. A platform subscription therefore does not grant unrestricted use of third-party financial datasets.[21]

Direct sharing and listings form part of the service under the applicable sharing terms. Marketplace participation and on-platform monetisation are optional offerings with separate Provider and Consumer Terms and an acceptance process. Technical access to a listing and the commercial rights to use its content are distinct parts of the arrangement.[11]

For a Marketplace product, the listing provider is the seller of record and responsible for the product. Snowflake’s policy directs commercial disputes to that provider and requires provider authorisation for refunds. That allocation matters for financial datasets and applications purchased through the platform: product scope, content rights and supplier obligations remain identifiable alongside Snowflake’s infrastructure service.[12]

Under the customer Data Processing Addendum, Snowflake acts as processor or subprocessor for Customer Personal Data, following the agreed instructions. Its role for personal data in service-usage records is treated separately as controller. The addendum’s deletion provisions follow the applicable retrieval period; data-protection roles should not be conflated with ownership of the underlying dataset.[22]

Snowflake Inc. is a Delaware corporation listed on the New York Stock Exchange under SNOW. Its current contact page identifies Menlo Park, California, as headquarters and lists 135 Constitution Drive.[27][25]

The entity table is determined by the customer’s Ship To Country in the agreement or order form. It is a separate choice from the cloud and region hosting an account. Contractual notices and the relevant regional addenda follow the applicable legal entity.[20]

Snowflake acquired Crunchy Data in June 2025. That business brought an enterprise Postgres distribution, Kubernetes tooling and the managed Crunchy Bridge service; the team now contributes to Snowflake Postgres. The acquisition is provider-family context, while individual deployment models and product terms retain their specific scope.[29]

The annual report records completion of the Observe acquisition on 2 February 2026. Observe adds observability technology for telemetry and troubleshooting to Snowflake’s portfolio. It is treated here as an acquired business, distinct from a data-provider partnership or a customer deployment.[27]

Snowflake’s legal resources publish separate terms for Crunchy Bridge, optional offerings, AI features and model/service pass-through obligations. Ownership under the same company does not make every acquired product or external model part of a single standard subscription. The selected offering identifies which additional terms apply.[33]

Operator

Snowflake Inc. and applicable regional Snowflake contracting entities

Service scope

Managed cloud data infrastructure, analytics, governance and AI for institutional data workflows.

Customer types

Asset managers; Wealth managers; Financial institutions; Data providers; Application developers

Delivery

SaaS; Hosted; API

Pricing publication

Public and quoted pricing

Headquarters

Menlo Park, California, United States

Platforms

Snowflake data platform; Snowpark; Cortex AI for Financial Services; Horizon Catalog; Snowflake Postgres

Integrations

AWS, Azure and Google Cloud accounts; Snowpipe and Snowpipe Streaming; Snowpark Python, Java and Scala; Iceberg REST Catalog API; Snowflake-managed MCP server; Secure Data Sharing and cross-region replication

Pricing scope

Public usage rates and edition choices; capacity, support and optional products depend on their agreements.

Implementation

Select account cloud and region, configure roles and data ingestion, and set workload and retention controls.

Support

Included Snowflake Support and contracted Priority Support, with severity-based initial-response targets.

Permissions and regulatory scope

Data infrastructure and analytical software; data entitlements and financial-service permissions remain with the relevant suppliers and institutions.

Provider metadata checked 21 Oct 2026Source 25Source 13Source 6Source 4Source 14Source 23Source 1Source 17Source 20Source 22Source 2Source 7Source 31Source 9Source 26Source 19Source 29Source 30Source 3Source 28Source 33Source 12Source 32Source 24Source 10Source 5Source 18Source 8Source 21Source 16Source 27Source 15Source 11

On 15 October 2026, Snowflake announced general availability of data mirroring for Snowflake Postgres. It automatically carries inserts, updates, deletes and schema changes into analytical Snowflake tables, across the supported AWS and Azure Postgres regions. The announced route is one-way from Postgres to Snowflake; the article describes the reverse direction as future work. Transactional and analytical services remain distinct even where their integration is managed.[31]

State Street is a named production example. Snowflake’s case study describes State Street building its Alpha Data Platform on Snowflake and Microsoft Azure, with client and third-party data sharing and intraday portfolio information for cash, collateral, securities lending and risk. This is the vendor’s account of an implemented platform, rather than merely a selection announcement. It does not make State Street’s custody or investment services a Snowflake product.[32]