Updated 5 Aug 2026 • 3 mins read
Amazon ElastiCache Pricing Explained (2026)
Amazon ElastiCache pricing has two models: node-based, billed per node-hour, and serverless, billed per GB-hour of stored data plus ECPUs. Engine choice matters too, with Valkey about 20-33% cheaper than Redis OSS. This guide explains ElastiCache pricing in 2026 and how to reduce it.
Most teams meet ElastiCache pricing twice. The first time, they spin up a cache, it works, and the bill looks small. The second time, months later, they notice a dozen staging caches that were never deleted, a production cluster sized for a traffic peak that lasts two hours a day, and a Redis engine they could have swapped for a 20 percent cheaper one on day one. ElastiCache is not expensive; it is just full of choices that quietly compound.
This guide explains Amazon ElastiCache pricing in 2026 clearly: the two billing models, how the three engines differ on price, where reserved nodes and Database Savings Plans fit, and the levers that bring the bill down.
The short version ElastiCache bills two ways. Node-based pricing charges a fixed hourly rate per node (from about $0.0128/hour for a cache.t4g.micro on Valkey). Serverless charges for data stored per GB-hour plus compute in ECPUs, with a minimum floor even when idle. Valkey is roughly 20 percent cheaper than Redis OSS on nodes and about 33 percent cheaper on serverless. Reserved Nodes cut node cost up to 55 percent; Database Savings Plans cut serverless up to 30 percent.
The Two Pricing Models
Everything in ElastiCache pricing starts with one decision: node-based or serverless.
Node-based pricing
You choose an instance size and pay a fixed hourly rate per node, and the meter runs whether the cache is busy or idle, from the moment the node reaches the Available state until you delete it. Partial hours bill as full hours. The cheapest node is a cache.t4g.micro running Valkey at about $0.0128 per hour, roughly $9.34 a month, scaling up through memory-optimized r-series nodes for large datasets. Node-based is predictable and cheaper for steady, well-understood traffic, and it unlocks features serverless does not, including Global Datastore, data tiering, and Reserved Nodes.
Serverless pricing
Serverless removes capacity planning and charges two meters at once: data stored, billed per GB-hour, and compute, billed in ElastiCache Processing Units (ECPUs). An ECPU covers both vCPU time and data transferred; a simple read or write consumes 1 ECPU per kilobyte, and heavier commands consume proportionally more. On Valkey, storage is about $0.084 per GB-hour and compute about $0.0023 per million ECPUs; Redis OSS and Memcached run higher at about $0.125 per GB-hour and $0.0034 per million ECPUs. There is no scale-to-zero: a minimum storage floor is always billed, so serverless shines for spiky, unpredictable workloads and is wasteful for steady ones that would keep a node busy.
Engine Choice: Valkey vs Redis OSS vs Memcached
Since AWS added Valkey, an open-source, Redis-compatible engine stewarded by the Linux Foundation, engine choice became a real pricing lever. AWS prices Valkey about 20 percent below Redis OSS on nodes and about 33 percent below on serverless, and AWS now recommends Valkey for new deployments. The serverless minimum difference is dramatic: Valkey starts at a 100 MB floor, roughly $6 a month, while Redis OSS and Memcached start at a 1 GB floor, roughly $60 to $91 a month. Across a dozen staging environments, that 10x gap on idle dev caches adds up fast. Reserved Redis OSS nodes automatically apply to Valkey nodes in the same family and Region after an upgrade, so switching is low-risk.
| Engine | Node pricing vs Redis OSS | Serverless minimum | Notes |
|---|---|---|---|
| Valkey | ~20% cheaper | 100 MB (~$6/month) | AWS-recommended default for new caches |
| Redis OSS | Baseline | 1 GB (~$60–91/month) | Mature and widely used |
| Memcached | Similar to Redis OSS | 1 GB | Simple cache with no persistence or replication |
Discounts: Reserved Nodes and Database Savings Plans
For steady workloads, commitments cut the rate sharply. Reserved Nodes trade a 1-year or 3-year commitment for up to 55 percent off node-based pricing, and are the best discount vehicle for stable production clusters; on a six-node r7g cluster, 3-year all-upfront reserved can save roughly $14,500 a year versus on-demand. Reserved Nodes are not available for serverless. For serverless, the discount path is AWS Database Savings Plans, launched in December 2025, which cut serverless costs up to 30 percent for a 1-year commitment against a steady spend floor. Choosing Graviton-based nodes (r7g, m7g) with reservations compounds the savings, commonly 40 to 60 percent below on-demand x86.
How to Cut Your ElastiCache Bill
- Default to Valkey for new caches; the 20 to 33 percent discount is free performance-equivalent savings.
- Match the model to the workload: nodes for steady traffic, serverless for spiky or unpredictable load.
- Right-size nodes to sustained load, not the two-hour daily peak, and scale with replicas or serverless instead of oversizing.
- Buy Reserved Nodes for stable production clusters and Database Savings Plans for steady serverless spend.
- Delete idle and staging caches; there is no pause, so an Available cache bills until you delete it. Watch cross-AZ data transfer and backup storage too.
A FinOps platform makes the neglected part automatic, flagging idle and staging caches, oversized nodes, and reservation gaps before a cost review does. See our best FinOps tools roundup and our AWS cost optimization with FinOps guide for how that discipline works in practice.
What to Remember
ElastiCache rewards two early decisions and punishes neglect of a third. Pick Valkey unless you have a specific reason not to, and you save a fifth of the bill before you cache a single key. Match node-based to steady traffic and serverless to spiky traffic, and you stop paying for capacity you do not use. And then actually watch the caches you create, because the meter starts at Available and stops only at delete, with no pause in between. Do those three things and ElastiCache stays the cheap, invisible speed layer it is meant to be, rather than a line item someone discovers during a cost review six months too late.
FAQs
Two ways. Node-based pricing charges a fixed hourly rate per node. Serverless charges for data stored per GB-hour plus compute in ECPUs, with a minimum storage floor. Engine choice, node size, replicas, region, backups, and data transfer all affect the total.
For tiny or spiky workloads, Valkey Serverless starts at about $6 a month (a 100 MB floor). For steady traffic, the cheapest node is a cache.t4g.micro on Valkey at about $0.0128 an hour, roughly $9.34 a month, and Reserved Nodes cut node cost further.
Yes. AWS prices Valkey about 20 percent below Redis OSS on node-based clusters and about 33 percent below on serverless, and recommends it for new deployments. Valkey is a Redis-compatible open-source engine, so migration is typically low-risk.
An ElastiCache Processing Unit is the serverless compute billing unit covering vCPU time and data transferred. A simple read or write consumes about 1 ECPU per kilobyte, and heavier commands consume proportionally more, so request patterns drive serverless compute cost.
No. Reserved Nodes apply only to node-based clusters, where they save up to 55 percent for a 1-year or 3-year commitment. For serverless, use AWS Database Savings Plans, which save up to 30 percent on a 1-year commitment.
Node-based is usually cheaper for steady, predictable traffic that keeps a node busy. Serverless is cheaper for spiky or unpredictable workloads where a provisioned node would sit idle much of the day, since you pay only for stored data and requests above the minimum floor.