Google Cloud wants coding agents to be able to spin up a real database as easily as they create a file. On November 4, 2026 the company announced the preview of AlloyDB Express, a new edition of its AlloyDB for PostgreSQL service built for developers and AI coding agents that need short-lived databases for tests, prototypes and internal apps.
The pitch is speed and zero idle cost. Google says an AlloyDB Express database is ready in under two seconds and can be created through the gcloud CLI, the REST API, MCP, Terraform or the Cloud console. After 15 minutes without activity the database pauses on its own and stops charging for compute, then wakes up within seconds when the next query arrives. Each instance is small, with one to two vCPUs and 20GB of storage, according to Google's Agentic Data Cloud overview.
Pricing follows that usage pattern. Active compute costs $0.115 per compute-unit hour, billed by the second; paused compute costs nothing; and storage is $0.30 per GB-month for the space data actually uses. Google adds a free tier of 50 compute units a month. When a prototype is ready for production, a one-click upgrade moves it to standard AlloyDB while keeping data, schemas, users and connection strings, and adds high availability, read replicas and disaster recovery.
Security is the part Google stresses for agent-created databases. Administrators define IAM, network and organization policies once, and every Express database created by a developer or an agent inherits them. Connections default to public DNS endpoints that hide IP addresses and block inbound traffic until it is allowed, private networking is available through Private Service Connect or private DNS, and access can be restricted with IAM, VPC Service Controls and native PostgreSQL roles. The idea is that an agent can give every test its own isolated database without anyone setting it up by hand.
The edition also carries AlloyDB's AI features. Google lists vector search with pgvector HNSW and its ScaNN index, so embeddings and agent memory can sit next to operational data, calls to Gemini models from SQL through functions such as ai.if, ai.rank and ai.generate, and natural-language-to-SQL generation in AlloyDB Studio.
AlloyDB Express was one of several Agentic Data Cloud updates published the same day. Google also announced direct Apache Iceberg access in AlloyDB, letting a single PostgreSQL query join live operational data with lakehouse tables in cloud storage, which it claims is up to 7x faster on aggregation queries than Amazon's comparable PostgreSQL-compatible service. It pointed to the recently announced PostgreSQL for agents in AlloyDB, a separate preview for production fleets of agents querying live data, and said cross-cloud interconnect and caching for AWS and Azure and BigQuery Graph are now generally available.
The launch is a preview, so service levels, regional availability and limits may still change, and Google's speed and cost claims have not been independently tested. The 7x Iceberg comparison is Google's own benchmark claim. The broader bet is clear, though: cloud providers expect a growing share of databases to be created and thrown away by agents rather than by people, and they are shaping provisioning, pricing and guardrails around that pattern.