Google Cloud Database Services Compared (2026 Guide)

Short answer

For a normal application database on Google Cloud, start with Cloud SQL (MySQL, PostgreSQL or SQL Server); it is the simplest and has the cheapest entry point.
Move up to AlloyDB for PostgreSQL workloads that need read pools, analytics on live data or AI features, and to Spanner only when you need horizontal write scaling or multi-region consistency.
Firestore is the serverless, pay-per-operation choice for documents; Bigtable is for very large, single-key, high-throughput data; Memorystore is a managed Valkey, Redis or Memcached cache.
BigQuery is an analytics platform, not an application database; pair it with an operational database rather than replacing one.

How we know: Research-based: each service's purpose, data model, availability figures, free tiers and headline prices were checked against Google Cloud's product pages, documentation, free-tier page and pricing pages on cloud.google.com on 8 October 2026. We have not run these services for this comparison, so no performance claims are made; where Google makes a claim about its own products, we attribute it.

Google Cloud database options at a glance

Google groups its database services into relational, non-relational and in-memory products, with BigQuery alongside for analytics. The table summarises what each Google Cloud database is for. All of them are managed services, meaning Google runs the servers, replication, patching and backups; they differ in data model, how they scale and how you pay.

Google Cloud database services: decision table (October 2026)
Service	Type	Engine or API	How it scales	Billing model	Best for
Cloud SQL	Relational	MySQL, PostgreSQL, SQL Server	Vertical; read replicas	Per vCPU, GiB of memory and provisioned storage	Standard application databases and migrations
AlloyDB	Relational	PostgreSQL-compatible	Vertical primary; read pools up to 20 nodes	Per vCPU and memory per node; storage on use	Heavy PostgreSQL, HTAP, AI with vectors
Spanner	Relational (also graph, key-value, search)	GoogleSQL or PostgreSQL dialect	Horizontal, across zones or regions	Per node-hour by edition, plus storage	Global, strongly consistent, very high scale
Firestore	Document (NoSQL)	Firestore API; MongoDB-compatible API	Automatic, serverless	Per document read, write and delete, plus storage	Mobile, web and serverless apps
Bigtable	Wide-column (NoSQL)	Bigtable API, HBase-compatible client	Add nodes; autoscaling	Per provisioned node-hour, plus storage	Time series, IoT, very large key-value data
Memorystore	In-memory	Valkey, Redis, Memcached	Shards and replicas	Provisioned capacity	Caching, sessions, leaderboards
BigQuery	Analytics platform	GoogleSQL, Python	Serverless	Per TiB scanned or reserved slots, plus storage	Reporting, BI and data science

Relational databases on Google Cloud: Cloud SQL, AlloyDB and Spanner

Google offers three relational services, which form a ladder of capability and cost. Most teams should start at the bottom and move up only when they hit a documented limit.

Cloud SQL

Cloud SQL is Google's fully managed service for MySQL, PostgreSQL and SQL Server. Each instance is a VM with network storage; Google handles backups, failover, patching and monitoring. Two editions (Enterprise and Enterprise Plus) trade price for availability: up to a 99.99% SLA including maintenance on Enterprise Plus. Google says more than 95% of Google Cloud's top 100 customers use Cloud SQL. Limits are those of a single primary server: up to 64 TB of storage and writes on one instance. See Cloud SQL pricing.

AlloyDB for PostgreSQL

AlloyDB is a PostgreSQL-compatible service built on a Google engine with separate compute and regional storage. A cluster has one primary and optional read pools, all sharing storage, plus a columnar engine that speeds up analytical queries on live data and AI extensions for vector search and model calls. It costs more per vCPU than Cloud SQL and suits larger PostgreSQL workloads. A downloadable edition, AlloyDB Omni, runs outside Google Cloud.

Spanner

Spanner is a distributed relational database that keeps ACID transactions and strong (external) consistency while scaling horizontally across zones and regions. It supports GoogleSQL and a PostgreSQL dialect, and Google says it powers products such as Search, Gmail and YouTube. Editions go up to a 99.999% availability SLA with multi-region configurations. It has no suspend mode and needs Spanner-aware schema design, so it is overkill for a typical regional application.

Non-relational databases: Firestore and Bigtable

Both are NoSQL, but they solve different problems: Firestore is for application documents with flexible queries and no capacity planning; Bigtable is for very high-throughput access by row key. Neither supports SQL joins in the way the relational services do; see SQL vs NoSQL.

Firestore

Firestore is a serverless document database: there is nothing to provision, it scales automatically, and you pay per document read, write and delete plus storage. Google lists 99.99% availability for regional and 99.999% for multi-region databases, strongly consistent queries and ACID transactions. It integrates with Firebase for mobile and web apps, and its Enterprise edition offers a MongoDB-compatible API so existing MongoDB drivers and tools can use it. The free quota (1 GiB of storage and 50,000 document reads a day) makes it the cheapest way to start a small app. See Supabase vs Firebase and Firebase alternatives.

Bigtable

Bigtable is a sparse, wide-column table that scales to billions of rows and thousands of columns, aimed at large amounts of single-keyed data read and written at low latency. Google recommends it for time-series, IoT, financial, marketing and graph-style data, with values typically no larger than 10 MB, and it offers an HBase-compatible Java client. You pay for provisioned nodes whether or not they are busy, so it only makes sense at scale. A free trial instance runs for 10 days, extendable to 90.

In-memory: Memorystore for Valkey, Redis and Memcached

Memorystore is Google's fully managed in-memory service for Valkey, Redis and Memcached, offering sub-millisecond data access. Memorystore for Valkey supports versions 7.2, 8.0, 9.0 and 9.1, in Cluster Mode Enabled (many shards) or Disabled (one shard), with up to five replicas per shard spread across zones. Memorystore for Redis offers a Basic tier and a Standard high-availability tier up to 300 GB with a 99.9% SLA. Use it as a cache, session store or leaderboard in front of Cloud SQL, AlloyDB or Spanner, not as the system of record. See Redis vs PostgreSQL.

Analytics: where BigQuery fits

BigQuery is Google's fully managed, serverless data platform for analytics, with separate storage and compute layers. It is built for scanning large data sets with SQL or Python, not for the many small reads and writes of an application. On-demand queries are billed by data processed, with the first 1 TiB each month free, or you can reserve capacity in slots. A common Google Cloud design keeps operational data in Cloud SQL, AlloyDB or Spanner and loads or queries it in BigQuery for reporting. For comparisons, see BigQuery vs PostgreSQL, Snowflake vs BigQuery and BigQuery alternatives.

How to choose a GCP database

In our assessment, four questions settle most choices:

Is the workload analytics or an application? Analytics over large data goes to BigQuery. Everything below is about application databases.
Relational or document data? If you need SQL, joins and transactions across tables, choose a relational service. If your data is naturally documents and you want zero capacity planning and pay-per-operation billing, choose Firestore. If you have massive single-key data, consider Bigtable.
How big, and how global? If one primary server can carry your writes and one region is enough, use Cloud SQL. If you are on PostgreSQL and need more read capacity or analytics on live data, use AlloyDB. If you need writes to scale out or strong consistency across regions, use Spanner.
Do you need a cache? Add Memorystore in front of whichever database you choose.

If you are comparing clouds rather than services, see AWS RDS vs Google Cloud SQL, Azure SQL vs Google Cloud SQL and Google Cloud SQL vs AlloyDB.

Pricing

Prices from the provider's official pricing pages, checked 8 October 2026, region Iowa (us-central1), in USD, excluding tax. List prices only; discounts, commitments and your actual usage change the bill.

Each service is billed differently, so the figures below are entry-level list rates, not comparable monthly costs. The detailed guides contain worked monthly estimates. Google's free features page lists free trials for Cloud SQL (30 days), AlloyDB (30 days) and Spanner (90 days, 10 GB), and always-free monthly allowances for Firestore and BigQuery. Memorystore prices are not summarised here.

Headline Google Cloud database list prices, Iowa (us-central1), USD
Service	Headline on-demand rate	Free allowance or trial
Cloud SQL (Enterprise)	0.0413 per vCPU-hour and 0.007 per GiB-hour of memory; db-f1-micro 0.0105 per hour	30-day free trial instance
AlloyDB (N2 and C4A)	0.06608 per vCPU-hour and 0.0112 per GiB-hour of memory	30-day free trial cluster
Spanner (regional, Standard)	0.90 per node-hour including three replicas (100 PUs = 0.1 node)	90-day free trial instance, 10 GB
Bigtable (Enterprise edition)	0.65 per node-hour	10-day free trial, extendable to 90 days
Firestore (Standard edition)	0.03 per 100,000 document reads; 0.09 per 100,000 writes	1 GiB storage, 50,000 reads, 20,000 writes and 20,000 deletes per day
BigQuery (on-demand queries)	6.25 per TiB processed	First 1 TiB of queries and 10 GiB of storage per month

Frequently asked questions

What is the main database service on Google Cloud?

For most applications it is Cloud SQL, the managed service for MySQL, PostgreSQL and SQL Server. Google states that more than 95% of Google Cloud's top 100 customers use it. AlloyDB, Spanner, Firestore and Bigtable cover workloads Cloud SQL is not designed for.

Which Google Cloud database is cheapest to start with?

Firestore and BigQuery have always-free daily or monthly allowances, so a small app or occasional analytics can run at no cost. Among relational services, Cloud SQL's shared-core machines are the lowest-cost option, and Cloud SQL, AlloyDB and Spanner all offer free trial instances.

Is BigQuery a database?

It stores data and answers SQL queries, but Google positions it as a serverless data and analytics platform. It is designed for large scans and reporting, billed by data processed or by reserved capacity, rather than for the frequent small transactions of an application.

Does Google Cloud offer a MongoDB-compatible database?

Yes. Firestore with MongoDB compatibility, part of Firestore Enterprise edition, lets existing MongoDB application code, drivers and tools use Firestore. Google's databases page also lists MongoDB Atlas as a partner service on Google Cloud.

Is there a Google Cloud database as a service for SQL Server?

Yes. Cloud SQL for SQL Server runs SQL Server 2017 to 2025 as a managed service, with a per-core licence included in the price. Cloud SQL does not accept licences you already own.

Which Google Cloud database is serverless?

Firestore and BigQuery are serverless: there is nothing to provision. Cloud SQL, AlloyDB, Spanner, Bigtable and Memorystore all require you to choose capacity (machines, nodes or processing units), although Spanner and Bigtable can autoscale.

Sources

Checked 8 October 2026.

How we research cloud database guides: our editorial method. CodeWithSQL earns nothing from the providers mentioned.