.png)
ElastiCache Serverless is the auto-scaling, no-capacity-planning configuration of Amazon ElastiCache. You create a cache in under a minute, and ElastiCache handles scaling automatically — from near-idle to hundreds of thousands of requests per second — without you choosing node types, cluster sizes, or shard counts.
The billing model is pay-per-use: you pay for ECPUs (ElastiCache Processing Units, a unified metric for compute and data transfer) and for GB-hours of data stored. Two billing dimensions, no instance charges.
The most important decision in ElastiCache Serverless is engine choice. Valkey Serverless is 33% cheaper per ECPU and per GB-hour than Redis OSS Serverless, and has a 100 MB minimum storage floor versus 1 GB for Redis OSS. For new deployments, Valkey is the recommended engine.
See exactly what you’re overpaying in under 60 seconds. Try the Calculator for free →
ElastiCache Serverless Pricing: Complete Rate Table
| Component | Rate (US East) | Notes |
| ECPU — Valkey Serverless | ~$0.0034/1,000 ECPUs | 33% lower than Redis OSS. 1 ECPU = 1 KB data transferred. Complex commands consume more ECPUs. |
| ECPU — Redis OSS Serverless | ~$0.0034/1,000 ECPUs (higher than Valkey) | Verify exact rates at aws.amazon.com/elasticache/pricing/ — Valkey is 33% lower. |
| ECPU — Memcached Serverless | Same as Redis OSS | No Valkey discount available for Memcached. Memcached does not support persistence. |
| Data stored — Valkey Serverless | ~$0.125/GB-hr (33% lower than Redis) | Minimum 100 MB metered storage. At 100 MB continuous: ~$6/month floor. |
| Data stored — Redis OSS / Memcached Serverless | ~$0.125/GB-hr | Minimum 1 GB metered storage. At 1 GB continuous: ~$91/month floor. Source: Usage.ai live blog. |
| Minimum metered storage — Valkey | 100 MB (~$6/month) | 90% lower minimum than Redis OSS. Ideal for dev, staging, and small caches. |
| Minimum metered storage — Redis OSS / Memcached | 1 GB (~$91/month) | Even a near-empty Redis OSS Serverless cache costs ~$91/month in storage alone. |
| Reserved Nodes | Not available for Serverless | Reserved Nodes apply to node-based clusters only. See ElastiCache RI guide for node-based discounts. |
| Database Savings Plans | Up to 30% savings | Covers ElastiCache Serverless on a 1-year commitment. Listed on AWS ElastiCache pricing page. |
| Scale-to-zero / pause | Not available | Billing starts when cache enters Available state. No way to pause a Serverless cache. Delete to stop billing. |
| Global Datastore | Not available for Serverless | Global Datastore is node-based only. Multi-region is not supported with Serverless. |
| Free tier (post July 15, 2025) | $100 in credits | New accounts get $100 credits (includes Serverless). Pre-July 2025 free tier excluded Serverless. |
All rates: US East (N. Virginia), June 2026. Source: AWS official pricing page and Usage.ai live ElastiCache blog. Verify at aws.amazon.com/elasticache/pricing/ — rates change.
ECPU rate ~$0.0034/1,000 ECPUs and storage ~$0.125/GB-hr from Usage.ai live ElastiCache blog (usage.ai/blogs/aws/reserved-instances/elasticache/). Minimum floors from AWS official pricing page. Verify all current rates at aws.amazon.com/elasticache/pricing/ — rates change.
What Is an ECPU?
An ElastiCache Processing Unit (ECPU) is the billing unit for ElastiCache Serverless. Each ECPU represents one kilobyte of data transferred per cache operation, bundling both vCPU time and data transfer into a single metric.
The ECPU formula: 1 ECPU per KB of data transferred. A GET command returning 3.2 KB consumes 3.2 ECPUs. A SET command writing 0.5 KB consumes 0.5 ECPUs. Commands that transfer more than 1 KB of data consume ECPUs proportionally. Source: AWS official pricing page — exact example cited.
Commands requiring additional vCPU time beyond the data transfer base also consume proportionally more ECPUs. Complex operations include: HSET with many field-value pairs, ZADD adding to large sorted sets, LRANGE returning large lists, Lua scripts executing server-side logic, and SORT operations. For most standard GET/SET workloads, ECPU consumption scales linearly with data transferred. For complex data structure operations, budget additional ECPUs beyond the 1-per-KB baseline. Source: Usage.ai live ElastiCache blog.
ECPU cost example: an API gateway caching JSON responses averaging 2 KB per response. At 1 million requests per hour: 1M x 2 KB = 2M ECPUs/hr. At $0.0034/1,000 ECPUs: 2M / 1,000 x $0.0034 = $6.80/hr = $4,964/month. The same throughput in a Valkey Serverless cache at 33% lower ECPU rate: $4,546/month. The difference of $418/month purely from engine choice.
The Valkey Advantage: 33% Cheaper Serverless, 15x Lower Minimum
AWS launched Valkey support for ElastiCache in October 2024. It is an open-source, Redis-compatible key-value store stewarded by the Linux Foundation and backed by over 40 companies. It is a drop-in replacement for Redis OSS — same commands, same protocols, same client libraries. Source: AWS Database Blog (October 2024).
The pricing advantage for Serverless is significant on two dimensions: 33% lower ECPU and storage rates, and a 100 MB minimum stored versus 1 GB for Redis OSS. That minimum difference alone determines the floor cost of your Serverless cache.
At minimum storage only (ignoring ECPU):
Redis OSS Serverless minimum: 1 GB x $0.125/GB-hr x 730 hrs = ~$91/month.
Valkey Serverless minimum: 0.1 GB x ($0.125 x 0.67)/GB-hr x 730 hrs = ~$6/month.
For a team running 10 development and staging caches at the minimum: Redis OSS = $910/month in storage alone. Valkey = $60/month in storage alone. The $850/month difference requires zero application changes — just engine selection at cache creation. Source: Usage.ai live ElastiCache blog and cloudburn.io (April 2026).
If you are running Redis OSS or Memcached Serverless today: existing Redis OSS reserved nodes automatically apply to Valkey nodes in the same instance family and region (for node-based clusters). For Serverless, switching to Valkey requires creating a new Serverless cache with the Valkey engine and migrating data. This is a one-time effort that pays off immediately in lower monthly bills. Source: AWS official ElastiCache pricing page.
Also read: ElastiCache Reserved Nodes: complete guide for node-based cluster discountsÂ
Serverless vs Node-Based: When Each Makes Sense
The ECPU model makes ElastiCache Serverless cost-effective for variable, bursty, and unpredictable workloads. For high-throughput production caches at consistent load, node-based clusters with Reserved Nodes deliver better unit economics.
At very high request volumes, Serverless costs can significantly exceed node-based costs. A node-based cache.r7g.xlarge (Valkey, US East) at approximately $0.350/hr is running roughly 1.26 million ECPUs per second of cache capacity continuously. If your actual request volume consistently demands that many ECPUs per second, the provisioned rate is cheaper than paying per ECPU. Source: Usage.ai live ElastiCache blog and cloudburn.io (April 2026).
| Scenario | Recommended | Reason | Approx saving |
| Dev, staging, test caches | Serverless (Valkey) | $6/month minimum vs $91 for Redis OSS Serverless or $30+ for smallest node. | Up to 95% vs over-provisioned node-based dev |
| New production cache, unknown traffic | Serverless | No capacity planning needed. Scale-out happens automatically. | Avoids over-provisioning at launch |
| Spiky workload (10x+ peak-to-average ratio) | Serverless | Pay for ECPUs consumed, not for peak capacity 24/7. | 30-60% vs node-based sized for peak |
| High-throughput production, consistent load | Node-based + Reserved Nodes | Fixed node-hour rate cheaper than per-ECPU billing at sustained high throughput. | Up to 55% vs Serverless at same sustained throughput |
| Multi-region replication required | Node-based with Global Datastore | Global Datastore not available with Serverless. | Serverless cannot serve this use case |
Source: cloudburn.io (April 2026), Usage.ai live ElastiCache blog, AWS official pricing page. Node-based Reserved Node savings from AWS official ElastiCache pricing (up to 55%).
Worked Examples: What ElastiCache Serverless Actually Costs
Example 1: Session Cache (Low Traffic)
A web application storing user session tokens. 10,000 active sessions, each 1 KB. Cache hit rate: 95%. 200,000 requests per hour (50 RPS average). Data stored: 10 MB. Engine: Valkey Serverless.
ECPU cost: 200,000 requests/hr x 1 KB average = 200,000 ECPUs/hr. At $0.0034/1,000 ECPUs: $0.68/hr = $496/month. Storage cost: Valkey minimum 100 MB metered = 0.1 GB x (effective per-GB-hr rate). Monthly storage at minimum: approximately $6/month. Total: approximately $502/month.
For this workload, a node-based cache.t4g.small (Valkey, ~$0.018/hr on-demand = $13/month) would be significantly cheaper at sustained load. The Serverless advantage is at lower RPS or highly variable traffic. Source: math from verified rates.
Example 2: Application Cache (Variable Traffic)
An e-commerce platform caching product catalog pages. Traffic: 10x higher during sales events (10am-6pm weekdays), near-idle overnight. Average: 50,000 ECPUs/hr off-peak, 500,000 ECPUs/hr during business hours. Data stored: 5 GB. Engine: Valkey Serverless.
Business hours (8 hr/day x 22 days = 176 hrs): 500,000 ECPUs/hr x $0.0034/1,000 = $1.70/hr x 176 = $299.20. Off-hours (730-176 = 554 hrs): 50,000 ECPUs/hr x $0.0034/1,000 = $0.17/hr x 554 = $94.18. Storage: 5 GB x (Valkey GB-hr rate) x 730 = approximately $61/month. Total: approximately $454/month.
For a similar consistently-loaded node-based cache.r7g.large Valkey (~$0.175/hr x 730 = $127.75/month), Serverless costs 3.5x more at sustained average load. The Serverless value here comes from the 10x peak-to-average ratio: the Serverless cache handles the sales event peaks automatically without pre-provisioning a cluster sized for 10x load 24/7. Source: math from verified rates.
Example 3: High-Scale Gaming Leaderboard
100 GB of sorted set data, 500,000 requests per second using ZADD and ZRANGE operations (complex sorted set commands consume significantly more ECPUs than simple GET/SET). Source: cloudburn.io (April 2026) citing AWS Serverless cost calculations.
Storage: ~$8.37/hr. ECPUs (complex sorted set operations at high RPS): ~$12.30/hr. Total: ~$20.67/hr = approximately $14,882/month. At this scale, a provisioned node-based cluster with Reserved Nodes would be significantly cheaper. This is the canonical example where Serverless is the wrong choice — high sustained throughput with complex operations.
See exactly what you’re overpaying in under 60 seconds. Try the Calculator for free →
Database Savings Plans for ElastiCache Serverless
Reserved Nodes are not available for ElastiCache Serverless. Source: AWS official ElastiCache pricing page: ‘Reserved nodes are not available with ElastiCache Serverless.’
The discount path for Serverless is Database Savings Plans. AWS lists Database Savings Plans as one of three pricing options for ElastiCache alongside on-demand and serverless. DSP provides up to 30% savings on a 1-year commitment for Serverless usage. Source: AWS official ElastiCache pricing page and Usage.ai live ElastiCache blog.
Note: DSP savings for ElastiCache Serverless (up to 30%) are lower than DSP savings for some other services (Neptune: up to 35%, Keyspaces: up to 35%). This reflects the different pricing structure of ECPU-based billing versus provisioned instance billing. Even at 30%, for a $500/month Serverless cache, DSP delivers $150/month = $1,800/year in savings on a 1-year no-upfront commitment.
AWS Database Savings Plans: Complete GuideÂ
How Usage.ai Handles ElastiCache Serverless Costs
Usage.ai monitors ElastiCache Serverless ECPU and storage consumption from CloudWatch and identifies three primary optimization opportunities: caches running Redis OSS where Valkey would reduce costs by 33%; caches at sustained high ECPU throughput where node-based clusters with Reserved Nodes would be cheaper; and Database Savings Plans commitment sizing for stable Serverless spend floors.
For engine migration from Redis OSS to Valkey: Usage.ai surfaces this as a specific recommendation for Serverless caches, quantifying the exact monthly saving at current throughput and storage levels. The 33% ECPU reduction and 15x lower minimum storage floor make this the highest-ROI optimization for most Redis OSS Serverless users who do not need Redis-specific features.
Usage.ai added native Database Savings Plans support in January 2026. For ElastiCache Serverless workloads with predictable minimum ECPU and storage spend, the platform recommends the DSP commitment level at the consistent floor of hourly spend. Fee: percentage of realized savings only.
Start your free ElastiCache savings analysis with Usage.ai
Frequently Asked Questions
1. What is ElastiCache Serverless?
Amazon ElastiCache Serverless is an auto-scaling, no-capacity-planning cache configuration. You create a cache in under a minute and ElastiCache scales compute and memory automatically. Billing is per ECPU (ElastiCache Processing Unit, covering vCPU time and data transfer) and per GB-hour of data stored. It supports Valkey, Redis OSS, and Memcached engines. For new deployments, Valkey is recommended — it is 33% cheaper per ECPU and per GB-hour, and has a 10x lower minimum storage floor. Source: AWS official pricing page and AWS Database Blog (October 2024).
2. How much does ElastiCache Serverless cost?
Two billing dimensions: ECPUs at approximately $0.0034 per 1,000 (Valkey; Redis OSS is 50% higher per ECPU) and storage at approximately $0.125/GB-hour (Redis OSS/Memcached; Valkey is 33% lower). Minimum floor: Valkey Serverless ~$6/month, Redis OSS/Memcached Serverless ~$91/month. At 50,000 RPS with 10 GB stored: approximately $900/month. At 500,000 RPS with complex sorted set operations (100 GB stored): approximately $14,882/month. Source: Usage.ai live blog, cloudburn.io (April 2026). Verify at aws.amazon.com/elasticache/pricing/.
3. What is an ECPU?
An ElastiCache Processing Unit measures the compute and data transfer consumed per cache operation. 1 ECPU = 1 KB of data transferred. A GET returning 3.2 KB consumes 3.2 ECPUs. Complex commands (HSET with many fields, ZADD to large sorted sets, Lua scripts) consume more ECPUs than their data transfer alone. ECPU rate: approximately $0.0034 per 1,000 ECPUs (Valkey). Source: AWS official ElastiCache pricing page.
4. Why is Valkey cheaper than Redis OSS on ElastiCache Serverless?
AWS launched Valkey in October 2024 at a 33% lower ECPU and storage rate than Redis OSS Serverless, and with a 100 MB minimum stored versus 1 GB for Redis OSS. Valkey is an open-source, Redis-compatible fork maintained by the Linux Foundation. The lower pricing is the permanent structure AWS launched with — it is not a promotional rate. For teams not using Redis-specific features absent from Valkey, switching to Valkey Serverless immediately cuts the bill by 33% with no application code changes. Source: AWS Database Blog (October 2024) and AWS official pricing page.
5. Can you use Reserved Nodes with ElastiCache Serverless?
No. Reserved Nodes are only available for node-based ElastiCache clusters. The discount path for ElastiCache Serverless is Database Savings Plans, which provide up to 30% savings on a 1-year commitment. For node-based clusters, Reserved Nodes provide up to 55% savings. Source: AWS official ElastiCache pricing page.
6. When is node-based ElastiCache cheaper than Serverless?
When sustained average ECPU consumption would cost more per hour than the equivalent node-based instance rate. At 50,000 RPS with 1 KB average payload: 50,000 ECPUs/second = 180M ECPUs/hr at $0.0034/1,000 = $612/hr for Serverless. The equivalent provisioned cache.r7g.xlarge Valkey at approximately $0.350/hr is 1,748x cheaper. Serverless is cheaper for variable workloads where average ECPU consumption is a fraction of sustained peak. Use CloudWatch to measure your actual average ECPU before deciding. Source: cloudburn.io (April 2026) and Usage.ai live blog.
