Overview
Google Cloud Spanner is the fully managed distributed relational database that originated as Google's internal storage system for Ads, Photos, and Play, and was opened externally in 2017. It combines horizontal scale, strong external consistency through Google's TrueTime API, and ANSI SQL — including a PostgreSQL-compatible interface added in 2022. Spanner is unique among commercial databases in achieving externally consistent transactions across regions on a planet-scale infrastructure.
The 2024 launch of granular Processing Units (PU) pricing — replacing the older Node-based model, with each PU equivalent to 1/1000 of a node — meaningfully lowered the entry cost. Spanner now competes directly with CockroachDB (cloud-portable distributed SQL), Aurora DSQL (AWS-native distributed SQL), and traditional Oracle RAC for tier-zero global workloads. The 2024 introduction of Spanner Graph (interleaved property graph queries) and built-in vector search expanded its applicability beyond the original OLTP scope.
Key Features
- External strong consistency across global regions via TrueTime API
- ANSI SQL with GoogleSQL and PostgreSQL-compatible interfaces
- Granular Processing Unit (PU) pricing — scale in 100 PU increments
- Multi-region configurations with read-only replicas in extra regions
- Spanner Graph for native property-graph queries (GQL)
- Built-in vector search with KNN and ANN indexes (HNSW)
- Spanner Vertex AI integration for in-database ML inference
- Change streams for capturing data changes into Dataflow, Pub/Sub, Kafka
- Point-in-time recovery within the configured backup window
- Online schema changes with no downtime for most operations
- Dual-region and multi-region configurations with documented 99.999% SLA
- Spanner Data Boost for elastic, separately-billed analytic queries
Pricing
| Configuration | Model | Cost (us-central1) |
|---|---|---|
| Compute (Regional) | Per PU/hour, 100 PU minimum | $0.30/PU/hour ($30/100PU/hour) |
| Compute (Multi-region eur) | Per PU/hour | ~$0.90/PU/hour |
| Storage (Regional, SSD) | Per GB/month | $0.30/GB/month |
| Storage (Multi-region) | Per GB/month | $0.50/GB/month |
| Backup storage | Per GB/month | $0.10/GB/month (Regional) |
| Network egress | Per GB | Standard GCP egress pricing |
| Committed Use Discount | 1-year or 3-year | 20% or 40% off compute |
Pricing verified May 2026 from Google Cloud public pricing pages. The 2024 introduction of PU-based pricing reduced the minimum entry cost by ~10x compared to the previous Node model. Multi-region pricing varies by configuration; check the specific config (nam-eur-asia1, etc.) for the exact rate.
Strengths
- True external consistency at planet scale — unique in commercial databases
- Operationally simple — no manual sharding, no leader election, no rebalancing windows
- 99.999% SLA on multi-region configurations is the strongest in the industry
- PostgreSQL interface meaningfully reduced adoption friction for new buyers
- Granular PU pricing lowered the entry cost dramatically
- Spanner Graph and Vector Search expand the addressable workload space
Limitations
- Cost remains high relative to single-region PostgreSQL for moderate workloads
- Vendor lock-in to Google Cloud is structural — migration off is non-trivial
- PostgreSQL interface has feature gaps (no triggers, limited PL/pgSQL, no extensions)
- Multi-region writes incur the TrueTime commit-wait latency tax
- Operational tooling and ecosystem narrower than Postgres or Aurora