The economics of Neon’s 2026 serverless Postgres pricing

Scaling economics and idle capacity

Neon undercuts Amazon Aurora for workloads that remain idle for significant periods because its architecture allows compute to scale to zero. Aurora Serverless v2 requires a minimum of 0.5 ACUs, which creates a minimum charge even when no queries run. In a performance testing scenario where the database remains idle 80 percent of the time, Neon compute costs $70.08 while Aurora Standard costs $87.60. Because Neon charges $0.106 per CU-hour on the Launch plan, while Aurora Standard charges $0.12 per ACU-hour, this pricing gap widens as usage increases. If you manage multiple ephemeral testing environments, you should consider how these minimum capacity charges accumulate across your AWS bill.

Aurora scales compute capacity in 0.5 ACU increments. While Aurora can scale up to high capacities, it cannot scale to zero. This differs from Neon, which suspends the database after a configurable idle period, such as the default five minutes. For an I/O-intensive application with a high compute requirement, Aurora I/O-Optimized costs $0.16 per ACU-hour. In this specific scenario, where a system is idle two-thirds of the time and uses high compute for the remaining third, Aurora compute costs $350.40 while Neon compute costs $311.46. If a developer manages a social media analytics platform with 511 off-peak hours and 219 low compute hours, Aurora Standard costs $219 while Neon costs $292 because they do not use the scale-to-zero functionality for these specific peak loads. Aurora I/O-Optimized is designed for I/O-intensive applications. It costs $0.16 per ACU-hour. In scenarios where there is no idle time, both Aurora I/O-Optimized and Neon charge $0.16 per ACU/CU-hour.

AI agent workloads and platform alternatives

Storage prices dropped to $0.35 per GB-month and compute costs fell by roughly 25 percent following the Databricks acquisition of Neon in 2025. According to company telemetry, 80 percent of Neon databases consist of AI agents rather than human developers. These agents require the instant provisioning and scale-to-zero capabilities that Neon provides via its shared-storage architecture. Supabase provides a different value proposition for teams needing a full backend-as-a-service. A Supabase Pro subscription costs $25 per month and bundles authentication, storage, edge functions, and realtime features.

Neon remains the superior choice for pure database workloads where bursty traffic patterns allow scale-to-zero to minimize expenses. For teams using a fleet of many small databases, Neon provides per-project allowances that stretch for prototypes. Each of up to 100 projects gets its own free 100 CU-hours and 0.5 GB per month. Supabase provides a Pro plan that includes 8 GB of database and 100 GB of storage. For users with high traffic, Neon’s Scale tier provides 50 GB of storage for $69 per month and allows autoscaling up to 8 CPU. Supabase Pro also includes 100,000 MAUs. Neon’s free tier allows up to 2 CU of autoscale.

Comparison of specifications and performance

Neon uses a disaggregated architecture that separates compute from storage. The compute nodes are stateless and connect to a Pageserver over the network. This separation enables the copy-on-write branching feature, which creates isolated database instances in under a second. This capability avoids the massive data duplication costs that traditional snapshotting methods impose on teams during heavy testing. However, Neon’s scale-to-zero functionality introduces a cold start latency of between 300 and 800 milliseconds when the compute node must initialize from an idle state.

Neon’s Launch plan provides up to 16 GB of RAM. The Scale tier provides 32 GB of RAM and allows autoscaling up to 8 CPU. Aurora storage scales up to 128 TB. Aurora Standard includes storage at $0.10 per GB-month and charges $0.20 per 1 million requests for I/O. Will the integration of AI agents into production workflows eventually make the cold start latency of serverless Postgres a dealbreaker?