Teradata Vantage

Overview

Teradata Vantage is the oldest lineage in this category and still the clearest answer to one specific question: what if you need a warehouse that does not have to run in someone else's cloud? It deploys on-premises as VantageCore or as a managed cloud service, and Teradata treats both as first-class.

The engine is shared-nothing massively parallel processing, built from virtual processors: parsing engines that compile and dispatch SQL, AMPs that execute work in parallel, and the BYNET interconnect between them.

Rows are distributed by hashing the primary index across AMPs, so adding nodes adds storage and processing together. That is a different mental model from the elastic warehouses, and index design still matters.

In the cloud the compute splits in two. Active compute serves always-on reporting and agentic workflows; elastic compute handles training, experimentation and seasonal bursts, and is spun up only when needed.

Elastic compute is the part that has modernised most: Teradata states it operates directly on open table formats, including Iceberg and Delta Lake, in low-cost object storage.

Lake access is proper read and write, through a DATALAKE object over external catalogues, reaching Unity Catalog, Glue, Hive Metastore, Iceberg REST including Polaris, and Gravitino, on S3, ADLS Gen2 or Google Cloud Storage.

Analytics are packaged as ClearScape Analytics: in-database functions spanning time series, anomaly detection and embeddings, bring-your-own-model and bring-your-own-LLM, ModelOps, and an enterprise vector store.

Temporal tables, including the ANSI valid-time forms, are the documented mechanism for history; there is no time-travel feature in the Snowflake sense.

Pricing is the weak point for comparison shoppers. Teradata publishes the model, a fixed-plus-flex commitment denominated in Teradata Units, but not a rate card, so budgeting needs a conversation.

Key features and capabilities

The same headings are used for every cloud data warehouses entry, so two tools can be read side by side.

Architecture and storage

• Shared-nothing massively parallel processing, built from parsing engines, AMPs and the BYNET interconnect
• Rows distributed by hashing the primary index, with AMP clusters for fault tolerance
• Cloud instances separate compute from storage for elasticity
• Active compute for always-on work, elastic compute for bursts and training

Performance and scaling

• Cloud instance sizes from S to 3XL, with instance counts from 2 to 128 on AWS
• Elastic compute provides on-demand capacity for experimentation and seasonal peaks
• Consumption is weighted by a logical IO ratio applied to instance rates
• Workload management, caching and auto-tuning detail is not published on the pages read

SQL and workloads

• Teradata SQL with documented ANSI and ISO SQL compliance
• Temporal tables, including ANSI valid-time forms, rather than a time-travel feature
• DDL for lake access through CREATE, ALTER, REPLACE and DROP DATALAKE
• JSON types, procedures and UDFs exist in the manuals but the specifics are not verified here

Open formats and sharing

• Read and write for Iceberg and Delta Lake, with analytics running directly on those tables
• Catalogues include Unity Catalog, AWS Glue, Hive Metastore, Iceberg REST, Polaris and Gravitino
• Object stores include Amazon S3, Azure Data Lake Storage Gen2 and Google Cloud Storage
• Lake data stays in your own object storage, and no egress rates are published

Security and governance

• Row-level security through security constraints on tables, capped at five constraints per table
• Security constraint columns and classifications administered through the security guide
• At-rest encryption documented for on-premises deployments
• Roles, access logging, private networking and named certifications are not verified on the pages read

AI and machine learning

• ClearScape Analytics covers time series, anomaly detection, natural language, embeddings and generative models
• An enterprise vector store for unstructured and vector workloads
• Bring your own LLM and bring your own model, with ModelOps for deployment
• Agentic AI through Teradata AI Studio and Teradata Fabric, including Tera and Tera Agents

Where it runs and what it costs

• Cloud on AWS, Azure and Google Cloud, with roughly 25 AWS regions listed
• On-premises through Teradata Factory on Dell hardware with GPU acceleration, IntelliFlex, or your own kit
• The only cloud warehouse here with a genuine on-premises path
• Consumption priced in Teradata Units under a fixed-plus-flex commitment

Pricing

Price on requestQuote only

Effectively quote-only. Teradata publishes the model but not a rate card: a fixed-plus-flex structure combining a base commitment with elastic capacity, denominated in a Teradata Unit that applies across compute, storage, software and AI, with two storage options and three packages, plus a calculator and a pricing specialist to talk to. The one published entry figure found is $4.80 an hour for VantageCloud Lake Standard, where units are consumed 20% more slowly than on the mid tier. A free ClearScape Analytics sandbox needs no card.

Vendor pricing page →

Demos and videos

• Exploring VantageCloud Lake Product demo · YouTube
• Teradata channel Channel or playlist · YouTube

About Teradata

Teradata Corporation is a public company trading on the New York Stock Exchange as TDC, founded in 1979 and headquartered at 17095 Via del Campo, San Diego, with offices including London, Sydney and Tokyo. Its own history describes a start in a Brentwood, California garage, which makes it older than every other product in this category by decades. The software is proprietary, sold as a subscription or consumption contract, with cloud usage denominated in Teradata Units that apply across compute, storage, software and AI. There is no open-source edition; openness comes through open table formats and catalogues.

Founded 1979 · San Diego, California · teradata.com