10 Best In-Memory Databases of 2026

1. 1

   Redis

   The original in-memory key-value data store, sold as Redis Cloud and Redis Software and downloadable as Redis Open Source.

   9.2overall

   The original in-memory key-value data store, sold as Redis Cloud and Redis Software and downloadable as Redis Open Source.

   Best forTeams that want the most widely used cache with a managed tier from $5 per month

   Adoption data

   28.3%In the 2026 Stack Overflow Developer Survey, 28.3% of respondents to the database question, or 3,414 developers, said they had done extensive work with Redis, the highest share of any in-memory store in the survey.

   Stack Overflow, 2026 Developer Survey: Databases (2026): “Redis 3,414 28.3%”

   Redis is the default answer for application caching and session storage, and the 2026 Stack Overflow survey backs that up: 28.3% of developers who answered the database question had worked with it, against 5.0% for Valkey. Every value lives in memory, keys can expire on a time-to-live, and LRU eviction drops cold keys when memory fills, which is the exact behavior a session store needs. Beyond plain strings it offers hashes, sorted sets for leaderboards, streams and geospatial indexes, plus replication, Sentinel failover and Redis Cluster for sharding. Redis Cloud starts with a free 30 MB database, then Essentials runs from $0.007 per hour, or $5 a month, on shared infrastructure with 250 MB to 100 GB. Pro adds dedicated deployments and active-active replication across regions, but carries a $200 monthly minimum. The open-source edition is now offered under AGPLv3, RSALv2 or SSPLv1, which matters if your legal team prefers BSD. We compared documented features and published prices.

   Strengths

   • +Free Redis Cloud database with 30 MB for prototypes
   • +Essentials plans cover 250 MB to 100 GB from $5 per month
   • +Used by 28.3% of database respondents in the 2026 Stack Overflow survey

   Limitations

   • −Pro plans carry a $200 monthly minimum
   • −Recent versions are not under the BSD license that Valkey keeps
   • −Redis Flex data tiering is needed to push large datasets beyond RAM

   SummaryRedis is the best choice for most caching and session workloads because it pairs rich data types and expiry with a managed tier that starts at $5 per month. Teams that require a BSD license should look at Valkey.

2. 2

   Valkey

   A BSD-licensed key-value datastore backed by the Linux Foundation, offered as a managed service by AWS, Google Cloud, Aiven and others.

   8.9overall

   A BSD-licensed key-value datastore backed by the Linux Foundation, offered as a managed service by AWS, Google Cloud, Aiven and others.

   Best forTeams that want a Redis-style cache under a permissive open source license

   Adoption data

   5.0%In the 2026 Stack Overflow Developer Survey, 5.0% of respondents to the database question, or 603 developers, said they had done extensive work with Valkey.

   Stack Overflow, 2026 Developer Survey: Databases (2026): “Valkey 603 5.0%”

   Valkey is the community fork that kept the Redis code base under the BSD license, and it now sits under the Linux Foundation, which the project says ensures it will remain open source. It includes 3-Clause BSD-licensed Redis code, and partners listed on valkey.io describe migrating freely between Redis and Valkey. The project's own homepage says the latest release, Valkey 9.1.3 from October 26, 2026, reaches 2.1 million requests per second on a single server thanks to a redesigned I/O threading model, and adds per-database ACLs. There is no license fee, and you can buy it as a managed service from Amazon ElastiCache, Aiven, UpCloud, Heroku and other providers listed on valkey.io, so you are not tied to one vendor. Adoption is younger: 5.0% of database respondents in the 2026 Stack Overflow survey had worked with it, against 28.3% for Redis. We compared documented features and published prices.

   Strengths

   • +No license fee and a permissive BSD license
   • +Several managed providers, so no single vendor lock-in
   • +Used by 5.0% of database respondents in the 2026 Stack Overflow survey

   Limitations

   • −No first-party managed service from the Valkey project itself
   • −Smaller share of developers than Redis
   • −Support comes from cloud providers or consultancies, not one vendor

   SummaryValkey is the best choice for teams that need a Redis-style cache under a BSD license because it costs nothing to run and is sold managed by several clouds. Pick Redis instead if you want one vendor for software and support.

3. 3

   Amazon ElastiCache

   AWS's managed caching service, with serverless and node-based options for Valkey, Memcached and Redis OSS.

   8.3overall

   AWS's managed caching service, with serverless and node-based options for Valkey, Memcached and Redis OSS.

   Best forAWS teams that want a serverless cache with no nodes to size

   Amazon ElastiCache is the shortest path to a managed cache for teams already on AWS. It runs Valkey, Memcached or Redis OSS, and AWS describes it as delivering microsecond latency for application caching and session stores. The serverless option removes node sizing: you pay for data stored and for ElastiCache Processing Units, where reads and writes use 1 ECPU per kilobyte transferred. AWS's own pricing example for US East shows Serverless for Valkey at $0.084 per GB-hour of data, and the minimum metered storage is 100 MB per Valkey cache, so a tiny cache costs a few dollars a month before requests. Node-based clusters suit steady loads, and Valkey 9.0 adds a durability option backed by a Multi-AZ transactional log. The catch is billing detail: request charges, data transfer and Extended Support premiums for older Redis OSS versions all add up. We compared documented features and published prices.

   Strengths

   • +Serverless for Valkey bills from a 100 MB minimum
   • +Runs Valkey, Memcached or Redis OSS behind one service
   • +Fits inside existing AWS networking, IAM and billing

   Limitations

   • −Requests are billed separately in ECPUs, 1 per kilobyte transferred
   • −New AWS accounts get credits rather than a permanent free tier
   • −Older Redis OSS versions pay an Extended Support premium of 80% or more

   SummaryElastiCache is the best choice for AWS teams that want a serverless cache because Valkey storage is billed at $0.084 per GB-hour with no nodes to size. Teams outside AWS get more for less from Redis Cloud or Upstash.

4. 4

   Dragonfly

   A multi-threaded in-memory data store that is API compatible with Redis, Valkey and Memcached, sold as Dragonfly Cloud.

   8.1overall

   A multi-threaded in-memory data store that is API compatible with Redis, Valkey and Memcached, sold as Dragonfly Cloud.

   Best forHigh-throughput caches that have outgrown a single-threaded Redis node

   Dragonfly targets teams whose Redis or Valkey bill keeps growing because each node uses only part of the machine. It is a multi-threaded, shared-nothing data store that speaks the Redis, Valkey and Memcached APIs, so existing clients connect without code changes. Dragonfly's own benchmark on its homepage shows 3,970,000 queries per second against 718,000 for Valkey and 148,000 for Redis, and lists 275 GB of memory used where Valkey used 400 GB; treat those as vendor numbers. Instacart and redBus are quoted on the same page describing smaller clusters and lower latency after moving. Pricing is clear for this category: the self-managed edition is free up to 100 vCPUs, Dragonfly Cloud's Flex tier starts at $36 per month on AWS, Google Cloud or Azure, and the Business tier with SSO, auto-scaling and a custom SLA starts at $2,000 per month. Adoption data is thinner than for Redis or Valkey. We compared documented features and published prices.

   Strengths

   • +Self-managed use is free up to 100 vCPUs
   • +Flex tier starts at $36 per month
   • +One process uses all cores instead of a cluster of single-threaded nodes

   Limitations

   • −Throughput figures come from Dragonfly's own benchmarks
   • −Business tier with SSO and custom SLA starts at $2,000 per month
   • −Not listed separately in the 2026 Stack Overflow database survey

   SummaryDragonfly is the best choice for teams with very large, busy caches because one multi-threaded instance replaces a Redis cluster and the cloud tier starts at $36 per month. Smaller apps will not need its throughput.

5. 5

   Memcached

   A free, open source distributed memory object caching system, first built for LiveJournal in 2003.

   7.9overall

   A free, open source distributed memory object caching system, first built for LiveJournal in 2003.

   Best forTeams that only need a simple, multithreaded key-value cache in front of a database

   Memcached is the plain cache in this list. The project describes itself as a free and open source distributed memory object caching system meant to speed up dynamic web applications by taking load off the database. Its about page explains the main idea: servers pool their spare memory so a given item is always stored and fetched from the same place across the cluster, which grows usable cache size as you add machines. That simplicity is the appeal. There are no data types to learn beyond keys and values, and little to tune. It is also the limit: the project site does not document persistence, replication, sorted sets or streams, so sessions vanish on restart and leaderboards need another tool. Memcached was built by Brad Fitzpatrick for LiveJournal in 2003 and is still offered by AWS as a managed ElastiCache engine. It was not listed separately in the 2026 Stack Overflow database survey, so we scored adoption conservatively. We compared documented features and published prices.

   Strengths

   • +No license fee
   • +Small, single-purpose design with little to configure
   • +Supported as a managed engine by AWS

   Limitations

   • −No persistence, replication or rich data types documented on the project site
   • −No first-party managed service
   • −Not listed separately in the 2026 Stack Overflow database survey

   SummaryMemcached is the best choice for teams that only need a fast object cache in front of a database because it is free and has almost nothing to configure. Use Redis or Valkey when sessions must survive a restart.

6. 6

   Upstash Redis

   A serverless, Redis-compatible database billed per request, with an HTTP REST API for edge and serverless functions.

   7.9overall

   A serverless, Redis-compatible database billed per request, with an HTTP REST API for edge and serverless functions.

   Best forServerless and edge apps with spiky or low traffic

   Upstash Redis is built for apps that run on serverless functions and edge runtimes, where a classic Redis connection pool is awkward. Alongside the normal Redis protocol it exposes an HTTP REST API, so a function can read a session or a rate-limit counter with a single HTTPS call. Billing follows usage: a free plan covers 250 MB of data, 10 GB of bandwidth and 500,000 commands a month, pay as you go costs $0.2 per 100,000 commands with data up to 100 GB, and fixed plans start at $10 a month for 250 MB with unlimited commands. Read regions can be added for global apps. For production you add the Prod Pack at $200 per month per database to get an uptime SLA, Multi-Zone high availability, encryption at rest and SOC 2. That makes Upstash cheap for small or bursty workloads and less attractive for large steady caches. We compared documented features and published prices.

   Strengths

   • +Permanent free plan for prototypes
   • +Price scales to zero when traffic stops
   • +Works from edge runtimes that cannot hold TCP connections

   Limitations

   • −Production features need the $200 per month Prod Pack
   • −Per-command billing can exceed fixed plans at steady high load
   • −Not listed separately in the 2026 Stack Overflow database survey

   SummaryUpstash Redis is the best choice for serverless and edge apps because it offers an HTTP API and bills $0.2 per 100K commands with a permanent free plan. Busy steady caches cost less on fixed-capacity services.

7. 7

   Aerospike

   A distributed NoSQL key-value database built for very large datasets with low latency, available as Community, Enterprise and Cloud editions.

   7.8overall

   A distributed NoSQL key-value database built for very large datasets with low latency, available as Community, Enterprise and Cloud editions.

   Best forLarge-scale user profile, adtech and fraud workloads that outgrow RAM

   Aerospike sits between a cache and a primary database. It is a distributed key-value store designed to keep latency low while datasets grow beyond what fits in memory, which is why its customer pages lean on adtech, fraud detection and user profile stores. Aerospike quotes Wayfair reaching 1 million transactions per second with sub-millisecond P99.9 latency on seven nodes, and Criteo peaking around 290 million queries per second; these are customer claims on the vendor site. The Community Edition is free and supports clusters of up to 8 nodes and 2.5 TB of data with two namespaces. Enterprise Edition lifts that to 256 nodes and no data limit, with strong consistency, transactions and multi-site clustering sold as additional licensing. Aerospike Cloud runs it as a managed service. Paid prices are not published, so we marked pricing as pending and scored value lower. We compared documented features and published prices.

   Strengths

   • +Free Community Edition covers sizable clusters
   • +Built to hold more data than fits in RAM
   • +Customer stories from PayPal, Criteo and Wayfair on aerospike.com

   Limitations

   • −No published USD prices for Enterprise or Cloud
   • −Multi-site clustering, strong consistency and transactions need extra licensing
   • −More setup than a single-node cache

   SummaryAerospike is the best choice for terabyte-scale profile and fraud data because its free Community Edition alone handles 8 nodes and 2.5 TB. It is more than a session cache needs.

8. 8

   Hazelcast

   A distributed in-memory computing platform for caching and stream processing, popular with Java teams.

   7.6overall

   A distributed in-memory computing platform for caching and stream processing, popular with Java teams.

   Best forJava and Spring teams that want a clustered cache with stream processing

   Hazelcast approaches in-memory data from the Java side. It is a distributed platform that implements JCache, the standard Java caching API, through client-server or cluster-only architectures, with integrations for Spring and Hibernate. Hazelcast lists distributed caching for sub-millisecond data access, geo and WAN replication for disaster recovery, and a choice between CP for consistency or AP for availability. It also processes streams, so a team can cache reference data and enrich events in the same cluster; BNP Paribas is quoted on hazelcast.com processing 10 million daily events with sub-120ms latency. Pricing is the weak spot for this list: Hazelcast Platform is licensed by annual subscription, Hazelcast Cloud is sold pay as you go or with prepaid credits, and both require contacting sales, so we marked pricing as pending. We compared documented features and published prices.

   Strengths

   • +Fits Java apps through JCache and Spring integration
   • +Combines caching and stream processing in one platform
   • +Self-managed or Hazelcast Cloud deployment

   Limitations

   • −No published USD prices
   • −Pricing requires a sales contact form
   • −Best suited to JVM stacks

   SummaryHazelcast is the best choice for Java and Spring teams because it implements the JCache standard and integrates with Spring and Hibernate. Teams outside the JVM have simpler and cheaper options.

9. 9

   Apache Ignite

   An Apache-licensed distributed database with memory-first storage, SQL and ACID transactions.

   7.5overall

   An Apache-licensed distributed database with memory-first storage, SQL and ACID transactions.

   Best forTeams that want SQL and transactions on top of an in-memory data grid

   Apache Ignite is a distributed database for workloads that need in-memory speed but also want SQL and transactions. Its site describes a memory-first design: data is served from RAM and can be persisted to disk, so a cluster restarts without reloading everything from another system. You can create tables, run SQL across distributed data and use ACID transactions, with clients for Java, .NET, C++ and Python. Session management and key-value stores are listed among its main use cases, next to event stream processing and feature stores. It is free under the Apache 2.0 license, and the latest Ignite 3 release, 3.1.0, came out on October 29, 2025. The trade-off is weight: Ignite is a cluster with data regions, eviction and persistence settings to tune, and there is no first-party managed service. It was not listed separately in the 2026 Stack Overflow survey. We compared documented features and published prices.

   Strengths

   • +Apache 2.0 license with no fee
   • +SQL over cached data instead of key lookups only
   • +Persistence lets data survive restarts

   Limitations

   • −No managed service from the Apache project
   • −Heavier to run and tune than Redis or Memcached
   • −Two major lines, Ignite 2 and Ignite 3, live in separate repositories

   SummaryApache Ignite is the best choice for teams that want SQL queries and ACID transactions over in-memory data because it offers both under a free Apache 2.0 license. It is too heavy for a simple session cache.

10. 10

    Momento

    A managed caching platform built on Valkey, with a serverless Cache Flex tier, dedicated clusters and pub/sub topics.

    7.3overall

    A managed caching platform built on Valkey, with a serverless Cache Flex tier, dedicated clusters and pub/sub topics.

    Best forTeams that want Valkey run for them with per-GB pricing

    Momento sells Valkey as a service and as tooling for teams that run it themselves. The managed side has two shapes: Momento Cache Flex, billed from $13 per GB-month with no cluster to size, and Momento Cache Cluster from $0.0105 per instance-hour for teams that want dedicated capacity. Topics adds pub/sub messaging from $1 per million operations. For self-hosting on Kubernetes, Momento prices a Valkey Operator at $3.50 per GB-month of managed capacity and an Enterprise Image at $1.50 per GB-month, with add-ons for data tiering and observability. Its homepage compares Valkey as-is at 239 thousand requests per second with 1 million requests per second after Momento's optimizations; that is the vendor's own test. The pricing page does not list a free plan, and adoption data is limited, so Momento ranks last here despite clear pricing. We compared documented features and published prices.

    Strengths

    • +Simple per-GB pricing for the managed cache
    • +Both managed and self-hosted Valkey options
    • +Built-in pub/sub for real-time messaging

    Limitations

    • −No free plan listed on the pricing page
    • −Performance figures come from Momento's own comparison
    • −Not listed separately in the 2026 Stack Overflow database survey

    SummaryMomento is the best choice for teams that want managed Valkey billed simply by stored data because Cache Flex starts at $13 per GB-month. Budget-minded teams with tiny caches will pay less on Redis Cloud or Upstash.

Questions or corrections: editorial@statlyst.com · Disclosure