Valkey 9 After 18mo: Redis Fork Reshapes Cloud Cache

Since October 2025, Valkey 9 has been handling over one billion requests per second in cluster deployments and has replaced Redis as the default option in major managed cache services. After 18 months, what began as a licensing fork has evolved into a technically independent project-and Redis 8 only partially catches up.

Key Takeaways

• Fork velocity: Just eight days after the Redis SSPL announcement in March 2024, Valkey was launched under the Linux Foundation-backed by AWS, Google, Oracle, Alibaba, Ericsson, Huawei, and Tencent (TechCrunch, 2024).
• Valkey 9 GA: Generally available since October 21, 2025. A cluster handles over 1 billion requests per second, with single-node throughput up to 40% higher than Valkey 8.1 (Linux Foundation, 2025).
• Cloud default: AWS ElastiCache and Google Memorystore have made Valkey the default for new instances. Valkey sees approximately 1 million container pulls per week (The Stack, February 2026).
• Redis 8 countermove: In May 2025, Redis Labs released version 8 with tri-licensing (RSALv2, SSPLv1, AGPLv3)-a return to open source, but under copyleft terms instead of BSD.
• Migration: According to maintainers, migrating from Redis 7.2 to Valkey is a “seamless patch.” AWS offers in-place upgrades with zero downtime.

18 Months of Valkey: From Protest Fork to Default

On March 20, 2024, Redis Labs switched from the BSD license to the Server Side Public License (SSPLv1)-a move aimed at hyperscalers offering Redis as a managed service without contributing back to the project. Just eight days later, Valkey emerged on GitHub as a fork of Redis 7.2.4, spearheaded by Madelyn Olson (AWS, formerly a Redis maintainer) and backed by the Linux Foundation.

What began as a licensing protest has, 18 months later, evolved into an independent codebase. Valkey 8.0 introduced I/O threading by the end of 2024; Valkey 8.1 accelerated TLS handshakes by roughly 300 percent. Valkey 9.0 reached general availability on October 21, 2025, featuring atomic slot migration, hash-field expiration, and multi-database support in cluster mode-none of which exist in Redis 8 in any comparable form.

The backing coalition is unusually broad: alongside AWS, Google, Oracle, Alibaba, Ericsson, Huawei, and Tencent appear as corporate sponsors on Valkey’s website. For an infrastructure project, this is a rare alliance-and a stark contrast to forks tied to a single vendor. Anyone familiar with the histories of OpenStack, MariaDB, or OpenTofu knows this well: multi-vendor governance under a neutral foundation is the single most important factor for a fork’s long-term stability. A lone provider can shift priorities at any time; a consortium of seven major tech players, however, embeds built-in checks and balances.

Under the hood, the Valkey team has also restructured code ownership differently than Redis did under its old model. Code reviews now run through a public maintainer board, and patches arrive from AWS, Google, and Alibaba engineers at roughly similar frequencies. This isn’t a guarantee of superior code quality-but it does provide a more stable foundation against unilateral strategic decisions. Those who wondered in 2024, upon the SSPL announcement, whether such a move could happen again will find in this governance structure the clearest answer the industry has offered.

Valkey 9 vs. Redis 8: Where the Real Differences Lie

The most common mistake in this debate is assuming it’s only about licensing. Technically, the projects have diverged since October 2025-and those differences impact production workloads.

Sources: valkey.io, redis.io, AWS Database Blog, Google Cloud Blog (2025-2026)

The key differentiator is atomic slot migration. Classic Redis Cluster resharding briefly generates redirect errors that manifest as client timeouts under high load. Valkey 9 migrates slots without downtime or client errors-a feature that makes a critical difference for multi-tenant cache clusters.

The Numbers Behind the Benchmark Dispute

Both projects tout dramatic performance leaps-and both sets of figures hold up on their own terms. Redis 8 reports up to 87% faster command execution and up to double the throughput compared to Redis 7.4. Valkey 9 claims up to 40% higher throughput versus Valkey 8.1-and for the first time pushes its cluster past the billion-requests-per-second threshold.

Anyone trying to directly compare these two numbers falls into a trap. Redis 8 benchmarks itself against its own prior version, 7.4, on a single-node basis. Valkey 9 measures cluster-wide scalability. The only reliable point of comparison comes from independent benchmarks-and so far, they confirm the Valkey trend: in the MayCoin benchmark, Valkey 8.0.1 achieved higher throughput and better p99 latencies than Redis 7.1.0 on identical hardware.

For teams making a decision today, the more meaningful indicator isn’t the percentage gain-it’s feature velocity. Valkey has delivered three major releases in just 18 months. Redis 8 marked a return to open source, but not yet proof of a sustainable release cadence.

Under the hood, Valkey 9 introduces three technical innovations that matter more in real-world benchmarks than the headline throughput figure. According to the release notes, memory prefetching for pipelined commands delivers roughly 40% higher throughput for batched operations. Zero-copy responses for large replies cut overhead by up to 20% by eliminating data copies between kernel and userspace. And SIMD optimizations for BITCOUNT and HyperLogLog accelerate certain analytics queries by up to 200%. Together, these optimizations target workloads Redis users have long identified as bottlenecks-analytics on bitfields, real-time aggregations, and large pipelines in batch jobs.

What Cloud Providers Are Doing Now

The clearest adoption curve comes from the hyperscalers. AWS launched ElastiCache for Valkey in 2024, making Valkey the default for new clusters and offering a managed upgrade path for existing Redis installations. Google Cloud has declared Memorystore for Valkey 9.0 generally available, while Oracle OCI has introduced its own managed service, OCI Cache. In all three cases, “Valkey as default” means users get Valkey-not Redis-unless they explicitly opt out.

Adoption metrics reinforce this trend. According to current Docker Hub data, Valkey logs around 1 million container pulls per week, totaling over 70 million since the fork. While this isn’t a direct measure of production usage, it’s a reliable indicator of integration into CI/CD pipelines, local development environments, and Kubernetes deployments. For a project just 18 months old, this level of integration is remarkable.

A real-world example from ElastiCache: Taiwanese crypto exchange MaiCoin performed a blue-green upgrade from Redis OSS to Valkey, benchmarking Redis 7.1.0, Valkey 7.2.6, and Valkey 8.0.1 under identical load conditions. The result? Valkey 8.0.1 delivered the highest throughput, lowest p99 latency, and best memory efficiency. For a production workload with strict latency requirements, these aren’t just marketing numbers-they translate directly into cost savings.

The DACH market is responding more cautiously-but understandably so. Larger managed Kubernetes environments on AWS Frankfurt and Google Cloud Frankfurt are now rolling out Valkey 9 for new clusters, while existing installations often remain on Redis 7.x. This aligns with the typical DACH pattern: licensing concerns are addressed first, and migration follows in the next capacity-planning cycle-rarely driven by panic, usually coordinated with release branches or scheduled cluster upgrades.

Migration: What “seamless patch” really means in practice

Valkey 9 speaks the Redis protocol (RESP) and supports Redis 7.2 clients without any code changes-making it the primary reason migrations often require zero application modifications. AWS ElastiCache provides a managed in-place upgrade path: new Valkey nodes are spun up within the existing cluster, replicate from the Redis primary, and then a failover switches traffic over. Downtime: none.

In reality, there are three pitfalls to watch for. First: Redis Stack modules like RedisJSON, RediSearch, or RedisBloom are proprietary and not included in Valkey. Users relying on them must either adopt Valkey’s community modules (Valkey-JSON, Valkey-Search, Valkey-Bloom-all licensed under BSD) or stick with Redis. While functional, these community modules don’t yet match the full feature set of their Redis Stack counterparts.

Second: Workloads depending on Redis Enterprise features such as Active-Active geo-replication won’t find an equivalent in Valkey. These use cases must remain on Redis or migrate to alternatives like Dragonfly or KeyDB.

Third: Observability tools that parse Redis-specific INFO fields may occasionally need adjustments. Major APM vendors (Datadog, Grafana, New Relic) now offer native Valkey support, but custom-built dashboards can still trip up.

When to stick with Redis-and when to switch

The decision is less technical than organizational. Teams already running Redis Enterprise or Redis Stack modules in production have no pragmatic migration path-and for most use cases, Redis 8 under AGPLv3 poses no legal issues, as long as the code isn’t resold as a service to third parties. The AGPL’s “viral” clause doesn’t apply to internal deployments.

By contrast, those using Redis purely as a classic key-value cache, relying on managed services from AWS or Google Cloud, and not requiring enterprise features will find Valkey 9 technically ahead today. Its license is more permissive, its feature velocity higher, and its cloud integration deeper. Moreover, the risk of a major cloud provider dropping support is far lower for a project backed by seven corporate sponsors under the Linux Foundation than for a solution tied to a single commercial vendor.

For teams starting fresh in 2026, the choice is already made: AWS ElastiCache and Google Memorystore now provision Valkey-not Redis-for new clusters. Choosing Redis requires an explicit commercial agreement.

Conclusion

After 18 months, the licensing debate has been settled-not because one side lost, but because both have evolved into stable options. Valkey 9 is now the technically more active, permissively licensed default among major cloud providers. Redis 8 has returned to open source under AGPLv3 and delivers performance gains, but it can’t keep pace with Valkey’s release cadence.

For most cloud workloads in the DACH region (Germany, Austria, Switzerland), Valkey 9 is today the lower-risk choice. Those migrating should start with a non-critical cache, evaluate Valkey community modules, and leverage their managed service provider’s in-place upgrade capabilities. The “seamless patch” works for 80 percent of use cases-the remaining 20 percent require planning time, not surprises.

Frequently Asked Questions

Source for header image: Pexels / Mikhail Nilov