AWS ElastiCache for Redis vs Memcached
ElastiCache is the AWS-managed offering for both engines. The interesting comparison is not the engines themselves (covered elsewhere on this site) but how AWS prices them, how cluster modes differ, and where the Valkey rebrand changes the maths.
The Valkey rebrand changed the pricing
In October 2024 AWS launched ElastiCache for Valkey at a 33% discount versus ElastiCache for Redis OSS. The announcement on the AWS blog explained the pricing as reflecting "the lower cost of running Valkey under the BSD license without the upstream Redis SSPL constraints." In practice this looks like the same node types (cache.r7g.large, cache.r6g.xlarge, etc.) listed at noticeably lower hourly rates for the Valkey engine than the Redis OSS engine.
For a cache.r7g.xlarge node (4 vCPU, 26GB RAM) in us-east-1, ElastiCache for Redis is roughly $0.302/hour on-demand, while ElastiCache for Valkey is around $0.201/hour for the same configuration (AWS pricing page, May 2026 figures, verify before procurement). Over a year of 24x7 operation for a small three-node cluster with one primary and two replicas, that is roughly a $2,800 versus $1,900 difference per year. Multiply by typical fleet sizes and the rationale for AWS pushing Valkey as the default becomes clear.
ElastiCache for Memcached has not seen a parallel discount because Memcached has always been BSD and the AWS pricing was never carrying a Redis-SSPL premium. The Memcached per-node prices have been roughly stable, broadly comparable to Valkey on equivalent node types. The headline ratio (Valkey is ~33% cheaper than Redis, Memcached is roughly equal to Valkey) means the engine choice for new ElastiCache clusters in 2026 is really Memcached versus Valkey on cost, with Redis OSS only making sense when you specifically need a Redis 8.0 feature that Valkey has not shipped yet.
Side-by-side: a typical cache.r7g.xlarge cluster
| Aspect | Memcached | Valkey | Redis OSS |
|---|---|---|---|
| Hourly per node (us-east-1) | ~$0.20 | ~$0.20 | ~$0.30 |
| Replication | None (scale via shards) | Up to 5 read replicas | Up to 5 read replicas |
| Persistence | None | RDB + AOF optional | RDB + AOF optional |
| Cluster Mode | Auto-discovery via config | Disabled or Enabled (500 shards) | Disabled or Enabled (500 shards) |
| Backup / Snapshot | Not available | Manual + automatic | Manual + automatic |
| In-memory data structures | Strings only | All Redis types | All Redis types + vector sets |
| Multi-threaded I/O | Yes (cores -1) | Yes (Valkey 8.x I/O threading) | Yes (Redis 6+ I/O threading) |
| TLS | Yes | Yes | Yes |
Source: AWS ElastiCache pricing, May 2026. Verify region pricing before procurement.
Cluster Mode: the architectural decision
ElastiCache for Redis and Valkey both offer two cluster modes: Disabled and Enabled. The names are misleading; both are clusters. Disabled means a single shard (one primary, up to five replicas, single point of failover); Enabled means a sharded keyspace across up to 500 nodes with cross-shard query support limited to operations within a single hash slot.
For most workloads, Cluster Mode Disabled is the right starting point. It is easier to operate, simpler for client libraries to integrate with, and supports any Redis command without the cross-slot restrictions. A single Disabled cluster on cache.r6g.16xlarge gives you 256GB of cache, plenty for many production use cases. The drawback is the lack of horizontal scaling: when 256GB is not enough, the migration to Enabled is non-trivial because keys may be hashed to a different shard and your client code must understand the cluster protocol.
ElastiCache for Memcached takes a different approach. There is no concept of replication, so scaling is purely horizontal: add more nodes, ElastiCache provides a config endpoint that exposes the list, and the client library does consistent hashing client-side. This is closer to the Memcached server-pool model of the early 2000s and works well if your application's client library supports the auto-discovery feature (most do). The catch is that node failures are visible to the application (the affected keys disappear from the cache and have to be repopulated from the data source).
The data-transfer and snapshot costs people forget
On-demand and reserved-node pricing is the visible cost of ElastiCache. The less visible costs are data transfer (free within the same VPC, charged when crossing AZs or regions) and backup storage. Multi-AZ deployments with cross-AZ replicas are billed for the cross-AZ traffic generated by the replication stream. For a busy cluster pushing 100MB/s of writes, that bill can run to a few hundred dollars a month per AZ pair before you do anything with the data.
Backup storage for Redis and Valkey clusters is free up to 100% of the cluster's reserved memory, then $0.085/GB-month beyond that. For a 256GB cluster keeping daily snapshots for two weeks, you can be looking at 3.5TB of backup storage and around $300/month if you exceed the free tier. Memcached clusters do not have backups, so this cost does not apply, but you also have no recovery story if the cluster goes down.
Reserved Nodes (1-year and 3-year, all-upfront / partial / no-upfront) reduce the visible per-hour cost by up to 55% for a 3-year all-upfront commitment. For predictable workloads this is a clear win; for elastic workloads the on-demand pricing avoids the lock-in. Savings Plans, which apply to EC2 and Lambda, do not apply to ElastiCache as of May 2026. Plan for the reservation up front if the workload is stable.
FAQ
Is ElastiCache Memcached cheaper than Redis?
Per-node, no. Both engines run on the same node types with the same per-hour price. The cost difference is architectural: Memcached has no built-in replication, so a cache.r7g.xlarge for Memcached gives you 26GB of cache, while the same node for Redis with one replica gives you 26GB of cache plus 26GB of replica capacity (and you pay for both). For pure ephemeral cache where you do not need replication, Memcached delivers double the cache per dollar.
What is ElastiCache for Valkey?
Launched October 2024, ElastiCache for Valkey is AWS's BSD-licensed fork of Redis 7.2.4. It is the default engine recommendation for new ElastiCache deployments since late 2024. Pricing is 33% lower than ElastiCache for Redis OSS on equivalent node types per AWS pricing page. Existing Redis OSS clusters can migrate in-place with a few clicks and zero data loss.
Cluster Mode Enabled vs Disabled?
Cluster Mode Disabled is a single shard with up to 5 read replicas. Easier to use, hard limit of one primary's memory (around 256GB on cache.r6g.16xlarge in 2026). Cluster Mode Enabled shards the keyspace across up to 500 shards (250 in earlier 2024 quotas) for horizontal scaling. Pick disabled until you outgrow it; the migration to enabled is non-trivial.
Does Memcached on ElastiCache support TLS?
Yes, since 2023. Memcached TLS support was added by ElastiCache before upstream Memcached had stable TLS, so it is an AWS-specific implementation. If you require TLS in your security baseline, both Redis and Memcached on ElastiCache satisfy it.
Reserved nodes and savings plans?
ElastiCache offers reserved nodes (1-year and 3-year, all upfront / partial / no upfront) at up to 55% discount versus on-demand. Same engine choices, same discount math. Savings Plans do not apply to ElastiCache as of May 2026; only Reserved Nodes.