Press Releases | Linux Foundation

Valkey 9.2 Lowers Cost of AI Workloads On Existing Infrastructure Alongside New Distributed KV Caching Capability

Written by The Linux Foundation | Nov 19, 2026, 6:00:00 PM

The latest release targets three of the largest costs of running at scale: extra memory held in reserve for backups, cross-region network traffic, and memory overhead for large datasets.

Summary

  • Valkey announced the general availability of version 9.2, introducing forkless snapshots, continuous stream compression, and B+ tree sorted sets to eliminate backup memory spikes, accelerate updates, and significantly reduce cross-region network costs.
  • Released alongside version 9.2, a new open-source distributed KV caching capability creates a shared cache across GPU fleets to reuse computed work, cutting AI inference costs by up to 40% and time-to-first-token by up to 80%. LMCache is the first open source inference project that will be involved in the capability.
  • By combining its well-known datastore with cost-saving features, Valkey 9.2 cements its position as a fast, high-performance engine for scaling modern AI workloads. Explore Valkey 9.2 here.

SAN FRANCISCO – November 19, 2026 – Valkey, the open source datastore for high-performing systems, today announced the general availability of Valkey 9.2, a release built to help organizations do more with the infrastructure they already have. Version 9.2 is complemented by the launch of distributed KV caching, taking aim at inference costs by providing a large, shared cache across entire GPU fleets. Building on Valkey’s core pillars of efficiency and performance, version 9.2 reinforces the project's commitment to scaling enterprise workloads while driving down infrastructure costs.

Valkey sits on the hot path of modern applications, powering the caching, queues, and real-time data that users immediately experience. At that scale, every gigabyte of memory and every request sent between regions has associated costs for businesses, whether it's a part of historical operations or a new AI initiative. Valkey 9.2 continues the project's commitment to being efficient by design.

“A major ethos of Valkey is doing more with less, which means getting more functionality with less hardware,” said Madelyn Olson, Valkey project maintainer and Principal Engineer for AWS in-memory databases. “We designed Valkey 9.2 around this principle to deliver reduced memory overhead, lower cross-region network costs, and worry-free performance. With each release, Valkey continues to prove its value as an exceptionally fast, efficient, and reliable datastore.”

Distributed KV Caching Maximizes GPU Servers and Minimizes Costs

Agentic AI applications waste expensive GPU compute by constantly reprocessing the same prompt instructions, tool definitions, and conversation history. Released alongside Valkey 9.2, open source distributed KV caching enhances AI workload efficiency by creating a large, shared cache across a GPU fleet, allowing inference engines to reuse computed work rather than recomputing. Builders running open-weight models on self-managed inference can now connect a distributed KV cache through LMCache, which plugs into engines like vLLM and SGLang using a single configuration change. Distributed KV caching achieves:

  • Lower AI inference spend: Reduces costs by up to 40% by allowing teams to serve more users on the GPUs they already have.
  • Improved latency: Decreases time-to-first-token by up to 80%.
  • Seamless deployment, no vendor-lock in: Works on any cloud or infrastructure with zero application code updates.

Path Hash Makes Finding AI Context More Efficient

While distributed KV caching saves expensive compute, efficiently locating that shared context has traditionally required engineering teams to build and maintain complex, custom lookup logic. Valkey 9.2 minimizes this burden with Path Hash, a new native data type that acts as a fast, lightweight index for AI caching. Path Hash is to built to:

  • Simplify lookups and architecture: With a single command users instantly find the longest matching prefix and exactly where its cached results are stored, keeping large AI data in GPU or host memory while Valkey serves as the fast, lightweight index.
  • Reduce engineering overhead: Custom routing logic for AI inference systems is replaced with a versatile, general-purpose building block to quickly find matches on any hierarchical data, extending beyond AI workloads.

Forkless Snapshots Are Backups Without the Memory Spike

Historically, taking database snapshots on busy systems caused massive memory spikes – sometimes doubling usage – forcing teams to over-provision and pay for idle memory just to handle backups. Valkey 9.2 prevents this costly penalty by introducing forkless snapshots, which save data using a background thread instead of a separate process. The feature supports:

  • A flat memory footprint: Snapshots eliminate the temporary memory spikes previously required to protect data and bring new replicas online.
  • Right-sized infrastructure costs and predictable performance: Operators size and pay for infrastructure based on real workloads rather than worst-case backup scenarios, ensuring consistent resource usage across the system.

Replication Compression Lowers Network Costs for Distributed Deployments

Valkey 9.2 also tackles expenses resulting from cross-regional data transfers by introducing continuous stream compression. Replication compression provides:

  • Decreased network and storage costs: Ongoing replication traffic is reduced by roughly 50% to 65% and backup file sizes shrink by up to 80% (using the built-in LZ4 option), directly lowering cloud transfer fees.
  • Faster recovery: New replicas synced about 35% faster, and ongoing replication traffic shrank by roughly 50% to 65%, at a cost of about 2% to 4% of a CPU core per replica.

Sorted-Set Configuration: Faster Rankings, Less Memory

Sorted sets power critical features like real-time rankings, gaming leaderboards, and priority queues, but their traditional structure consumes excess memory and slows down as datasets grow. Valkey 9.2 fixes this by rebuilding large sorted sets on a highly compact B+ tree architecture.

  • Significant performance and efficiency gains: Delivers up to 105% faster updates and 36-43% better memory efficiency for sets with millions of entries.

To see what else is included in Valkey 9.2 or upgrade to the latest version, visit valkey.io.

 

Supporting Quotes

“With representation on Valkey’s Technical Steering Committee, Alibaba Cloud is committed to advancing Valkey through open collaboration. We are proud to have contributed Path Hash to Valkey 9.2, bringing native support for hierarchical data and expanding what developers can build with Valkey. Drawing on our Tair team’s research and practical experience in KV cache for AI inference, we see significant opportunities for Valkey in AI infrastructure. We look forward to working with the community to help Valkey meet the evolving needs of AI workloads and deliver greater value to developers.”

– Guoyun Zhu, Head of Tair R&D, Alibaba Cloud

“Valkey has consistently delivered improved availability, performance and efficiency with each release. Valkey 9.2 is no exception, with forkless saves, hot key detection and even more efficient sorted sets. Momento and AWS have worked closely, and in the open, to advance RDMA for Valkey. It's a great example of the open source ecosystem working! Valkey proves that true open source doesn't lock innovation behind enterprise-license paywalls.”

– Khawaja Shams, CEO, Momento

“At Multiverse Computing, we help customers scale inference workloads with compressed models that cut cost and latency. We cut both even further and served more prompts on existing GPUs by adding distributed KV caching on Valkey. With Valkey, we processed 50% more requests on the same GPUs within our one-second target for median time to first token. We achieved a 23% reduction in our median time to first token at normal traffic, and a 95% reduction at peak traffic. We also contributed LMCache connector improvements to the open-source community for Valkey that make distributed KV-caching more resilient.”

– Ennio Pastore, Director of AI Platform, Multiverse Computing

“KV cache objects are large and latency-sensitive, which is exactly where traditional networking falls short. With RDMA, Valkey can move multi-megabyte KV chunks directly into GPU-serving nodes with minimal CPU overhead. That makes it a strong shared backend for LMCache deployments where many inference engines need to reuse the same context. More fast, open storage options means users can match their cache tier to their infrastructure instead of the other way around.”

– Yihua Cheng, CTO and Co-founder, Tensormesh

 

About Valkey

Valkey is the open source in-memory datastore built for the core workloads of modern applications: caching, session stores, message queues, leaderboards, and latency-sensitive data paths. Designed for operational simplicity and cost-optimized scale, Valkey consistently delivers low latency at high throughput, enabling teams to run critical data infrastructure at any scale, without vendor lock-in or licensing surprises.. Governed by the Linux Foundation, Valkey is backed by a rich ecosystem of contributors and industry participants, including Aiven, Alibaba Cloud, anynines, AWS, BetterDB, Bytedance, DigitalOcean, Ericsson, Google Cloud, Heroku, k0rdent, Momento, NetApp Instaclustr, Oracle, Percona, Tencent Cloud, and UpCloud.

 

Media Contact

The Linux Foundation
pr@linuxfoundation.org