Google Cloud Storage

Overview

Cloud Storage is Google Cloud's object store, organised into buckets of objects and billed on storage, operations, retrieval and network. Functionally it competes directly with S3 and Azure Blob, and the differences are in the details rather than the shape.

Consistency is its strongest published claim. Objects are immediately available for reading and metadata operations as soon as you get a success response, and that includes object listing, which is not universally true elsewhere.

The documented exceptions are narrow and worth knowing: access-control changes and bucket recreation take about a minute to settle.

Maximum object size is 5 TiB, and buckets ramp to roughly 1,000 writes and 5,000 reads a second before scaling further, with no published hard ceiling.

Hierarchical-namespace buckets are the newer option, giving real folders and up to eight times the queries per second of a flat bucket, at slightly higher per-operation prices.

There are five classes now that Rapid has joined. Rapid is optimised for I/O-intensive AI, ML and analytics work, with a 99.9% zonal service level agreement, and costs roughly four and a half times Standard.

Autoclass is the automatic tiering option, moving objects across Standard, Nearline, Coldline and Archive on 30, 90 and 365-day thresholds. Its real appeal is that retrieval and early-deletion fees do not apply while it is on.

For lakehouse work the pattern is BigQuery tables for Apache Iceberg, which keep data in buckets you own through BigLake and stay queryable by Spark and other engines.

Check the published limitations on those tables before designing around them: no renaming, cloning or snapshots, no materialised views or row-level security, and one concurrent mutating statement per table.

Interoperability with S3 is real but undocumented in detail. The XML API works with S3-compatible tools using HMAC keys, though Google publishes no feature-parity matrix, and gcloud can address S3 buckets directly.

Key features and capabilities

The same headings are used for every object storage entry, so two tools can be read side by side.

Storage model

• Flat-namespace buckets by default, with hierarchical-namespace buckets giving real folders
• Maximum object size 5 TiB, with object names up to 1,024 bytes
• Bucket creation and deletion limited to roughly one request every two seconds per project
• Strong global consistency for read-after-write, read-after-delete and listing

Performance and scale

• Buckets ramp to roughly 1,000 writes and 5,000 reads a second, then scale automatically
• Hierarchical-namespace buckets start at up to eight times that rate
• Rapid storage targets low latency and high throughput, with a 99.9% zonal service level agreement
• Rapid Cache is a billed read cache, and retrieval fees do not apply to data read from it

Storage classes and lifecycle

• Five classes, Rapid, Standard, Nearline, Coldline and Archive
• Minimum durations, none on Standard and Rapid, 30 days Nearline, 90 Coldline, 365 Archive
• Retrieval per GiB, free on Standard and Rapid, $0.01 Nearline, $0.02 Coldline, $0.05 Archive
• Autoclass tiers automatically for $0.0025 per 1,000 objects a month, and waives retrieval and early-deletion fees

Security and governance

• IAM permissions plus access control lists, with HMAC keys for S3-style authentication
• Bucket Lock enforces retention up to 100 years, and locking a policy is irreversible
• Google states Bucket Lock can help with FINRA, SEC and CFTC requirements
• No Cohasset assessment or SEC 17a-4 citation appears on the Bucket Lock page
• Customer-managed and customer-supplied encryption keys, with Cloud Audit Logs for access logging

Analytics features

• BigQuery tables for Apache Iceberg keep data in your own buckets through BigLake
• Those tables stay queryable by BigQuery and open-source engines including Spark
• Published limits include no cloning or snapshots, no materialised views, no row-level security
• There is no in-place SQL filter equivalent to S3 Select

Ecosystem

• The XML API interoperates with S3-compatible tools and libraries using HMAC keys
• No supported or unsupported S3 feature matrix is published
• gcloud can address S3 buckets directly, and Storage Transfer Service moves data from S3 and Azure
• Engine reach through BigQuery, BigLake, Dataproc, Spark and any Iceberg-compatible engine

Where it runs and what it costs

• 43 regions and 130 zones, with regional, dual-region and multi-region buckets
• Cloud Storage itself is cloud-only; Google Distributed Cloud publishes a separate S3-compatible API
• Standard storage $0.02 a GiB a month in Iowa, down to $0.0012 on Archive
• Egress charged from $0.12 a GiB, falling with volume

Pricing

Usage-basedStandard at $0.02 a GiB a month

In Iowa, storage runs $0.02 a GiB a month on Standard, $0.01 Nearline, $0.004 Coldline, $0.0012 Archive and $0.09 on the new Rapid class. Operations cost $0.005 per thousand for writes and $0.0004 for reads on Standard, slightly more on hierarchical-namespace buckets, and $0.05 either way on Archive. Retrieval is $0.01 to $0.05 a GiB on the colder classes, and egress starts at $0.12 a GiB. An always-free tier gives 5 GB-months in three US regions, 5,000 writes, 50,000 reads and 100 GB of North American egress a month.

Vendor pricing page →

Demos and videos

• Cloud Storage in a minute Product demo · YouTube
• What is smart storage? Google Cloud's storage vision Webinar or talk · YouTube

About Google Cloud

Cloud Storage is a first-party Google Cloud service with no separate ownership or funding. It began as Google Storage for Developers, with Google's own developer blog writing about it by June 2010, though an exact general availability date is not published on the current product pages. It is proprietary with no open-source or self-hostable edition, governed by the Google Cloud Platform terms of service and billed per Google Cloud billing account, with non-dollar currencies handled through Cloud Platform SKUs. Google publishes a footprint of 43 regions and 130 zones across six continents.

cloud.google.com