How Valkey Managed Database is intended to reduce latency and cost

Prompt and embedding caching

Caching can prevent an application from performing the same expensive work repeatedly. Exact prompt caching stores results for repeated requests, while semantic caching uses vector similarity to identify requests that are close enough to reuse an earlier result. In a typical semantic-cache flow, an application embeds a new query, searches for a similar cached request, and only calls the model when no valid match exists.

That can shorten response times and reduce the number of model calls for workloads with repeated or near-duplicate requests. The size of the benefit depends on the application’s cache-hit rate, freshness requirements, and model workflow; caching is not a universal replacement for retrieval or inference.

Native vector search for RAG

RAG applications convert documents and queries into embeddings, then search for nearby vectors before sending relevant context to a model. Valkey supports vector-search workflows, allowing an in-memory data store to handle key-value data, cached content, and vector retrieval in the same operational layer.

For some applications, that can simplify the retrieval path and reduce the need to operate a separate vector database. It does not mean every RAG system should abandon a specialized database: data volume, durability, filtering, indexing, and consistency requirements still determine the right architecture.

Persistent state for AI agents

Agents are typically multi-step systems. They may need to remember a conversation, retain tool results, track a task’s progress, and recover a checkpoint before continuing. Valkey’s in-memory data structures and vector-store capabilities can provide a fast working-memory layer for those operations; Valkey is also documented for session storage and other real-time workloads.

The practical distinction is between a stateless model request and a stateful application around that model. Valkey does not supply the agent’s reasoning or guarantee correct behavior, but it can provide the quickly accessible state that helps an application coordinate multiple model and tool calls.

What “managed” changes for enterprise teams

A self-operated Valkey deployment still requires teams to provision instances, configure replication and failover, monitor performance, apply updates, scale capacity, and plan for availability. Akamai’s managed offering is designed to shift those database responsibilities to the platform, allowing developers to consume Valkey as a service rather than build the operational layer themselves. Akamai describes the product as a fully managed, high-performance platform intended to reduce operational complexity and speed time to value.

That trade-off is important for enterprise AI teams. A managed database can reduce infrastructure work, but organizations still need to design data retention, cache invalidation, access control, durability, observability, and recovery policies around it.

How it fits with Akamai Inference Cloud

Akamai Inference Cloud is the compute and routing side of the company’s distributed-AI strategy. Launched in October 2025, it was designed to bring agentic AI inference to the edge, close to users and devices. Valkey Managed Database adds an adjacent memory and retrieval layer for the application state that inference needs.

The intended architecture is therefore broader than simply running a model at the edge:

1. Inference resources execute model workloads closer to users and data.
2. Valkey stores frequently accessed context, embeddings, cached results, and session state in memory.
3. Vector search helps retrieve semantically relevant information for RAG and semantic caching.
4. Managed operations reduce the work required to keep the data layer available.

Akamai has continued expanding the infrastructure behind this approach. It announced thousands of NVIDIA Blackwell GPUs for distributed AI workloads and described intelligent routing across its network as a way to reduce latency and data-egress issues associated with centralized data centers. Valkey does not replace that compute layer; it is meant to make the surrounding data path better suited to low-latency inference.

Why the launch matters to Akamai’s cloud strategy

The product broadens Akamai’s position from content delivery and security toward a more integrated distributed-cloud stack combining compute, data services, edge execution, and protection. The timing also reflects the growth of its infrastructure business: Cloud Infrastructure Services generated $99 million in second-quarter 2026 revenue, up 39% year over year.

Akamai’s first-quarter results separately cited a seven-year, $1.8 billion commitment from a leading frontier-model provider for Cloud Infrastructure Services. That commitment is evidence of growing infrastructure ambitions, but it should not be treated as a Valkey-specific contract or as proof that every element of Akamai’s distributed-AI roadmap is already broadly deployed.

Why Akamai chose Valkey

Valkey is an open-source, community-led in-memory key-value database that originated as a fork of Redis. The project is designed to remain open source, maintain compatibility with Redis protocols and data structures, and support caching and other real-time workloads.

For Akamai, Valkey offers a familiar high-performance data model for developers who already understand Redis-style caching and session storage, while aligning the service with an open-source ecosystem. The choice also gives the company a foundation for AI-specific uses such as semantic caching and vector retrieval, without presenting Valkey as a general-purpose replacement for every database category.

The bottom line

Valkey Managed Database is best understood as the memory and retrieval layer in Akamai’s distributed-AI architecture. Its value proposition is straightforward: keep hot AI data in memory, place it nearer to inference and users, reuse results when possible, support vector-based retrieval, and offload database operations to a managed service.

The strongest use cases are latency-sensitive assistants, RAG pipelines, semantic caching, and stateful agents. The launch is strategically significant because it connects Akamai’s edge-inference ambitions to the data infrastructure those applications require. But the product is still described as limited availability, and claims about exact global coverage, channel-partner availability, or specific future locations require additional product or partner documentation before they can be treated as confirmed.

How Valkey Managed Database is intended to reduce latency and cost

Prompt and embedding caching

Caching can prevent an application from performing the same expensive work repeatedly. Exact prompt caching stores results for repeated requests, while semantic caching uses vector similarity to identify requests that are close enough to reuse an earlier result. In a typical semantic-cache flow, an application embeds a new query, searches for a similar cached request, and only calls the model when no valid match exists.

That can shorten response times and reduce the number of model calls for workloads with repeated or near-duplicate requests. The size of the benefit depends on the application’s cache-hit rate, freshness requirements, and model workflow; caching is not a universal replacement for retrieval or inference.

Native vector search for RAG

RAG applications convert documents and queries into embeddings, then search for nearby vectors before sending relevant context to a model. Valkey supports vector-search workflows, allowing an in-memory data store to handle key-value data, cached content, and vector retrieval in the same operational layer.

For some applications, that can simplify the retrieval path and reduce the need to operate a separate vector database. It does not mean every RAG system should abandon a specialized database: data volume, durability, filtering, indexing, and consistency requirements still determine the right architecture.

Persistent state for AI agents

Agents are typically multi-step systems. They may need to remember a conversation, retain tool results, track a task’s progress, and recover a checkpoint before continuing. Valkey’s in-memory data structures and vector-store capabilities can provide a fast working-memory layer for those operations; Valkey is also documented for session storage and other real-time workloads.

The practical distinction is between a stateless model request and a stateful application around that model. Valkey does not supply the agent’s reasoning or guarantee correct behavior, but it can provide the quickly accessible state that helps an application coordinate multiple model and tool calls.

What “managed” changes for enterprise teams

A self-operated Valkey deployment still requires teams to provision instances, configure replication and failover, monitor performance, apply updates, scale capacity, and plan for availability. Akamai’s managed offering is designed to shift those database responsibilities to the platform, allowing developers to consume Valkey as a service rather than build the operational layer themselves. Akamai describes the product as a fully managed, high-performance platform intended to reduce operational complexity and speed time to value.

That trade-off is important for enterprise AI teams. A managed database can reduce infrastructure work, but organizations still need to design data retention, cache invalidation, access control, durability, observability, and recovery policies around it.

How it fits with Akamai Inference Cloud

Akamai Inference Cloud is the compute and routing side of the company’s distributed-AI strategy. Launched in October 2025, it was designed to bring agentic AI inference to the edge, close to users and devices. Valkey Managed Database adds an adjacent memory and retrieval layer for the application state that inference needs.

The intended architecture is therefore broader than simply running a model at the edge:

1. Inference resources execute model workloads closer to users and data.
2. Valkey stores frequently accessed context, embeddings, cached results, and session state in memory.
3. Vector search helps retrieve semantically relevant information for RAG and semantic caching.
4. Managed operations reduce the work required to keep the data layer available.

Akamai has continued expanding the infrastructure behind this approach. It announced thousands of NVIDIA Blackwell GPUs for distributed AI workloads and described intelligent routing across its network as a way to reduce latency and data-egress issues associated with centralized data centers. Valkey does not replace that compute layer; it is meant to make the surrounding data path better suited to low-latency inference.

Why the launch matters to Akamai’s cloud strategy

The product broadens Akamai’s position from content delivery and security toward a more integrated distributed-cloud stack combining compute, data services, edge execution, and protection. The timing also reflects the growth of its infrastructure business: Cloud Infrastructure Services generated $99 million in second-quarter 2026 revenue, up 39% year over year.

Akamai’s first-quarter results separately cited a seven-year, $1.8 billion commitment from a leading frontier-model provider for Cloud Infrastructure Services. That commitment is evidence of growing infrastructure ambitions, but it should not be treated as a Valkey-specific contract or as proof that every element of Akamai’s distributed-AI roadmap is already broadly deployed.

Why Akamai chose Valkey

Valkey is an open-source, community-led in-memory key-value database that originated as a fork of Redis. The project is designed to remain open source, maintain compatibility with Redis protocols and data structures, and support caching and other real-time workloads.

For Akamai, Valkey offers a familiar high-performance data model for developers who already understand Redis-style caching and session storage, while aligning the service with an open-source ecosystem. The choice also gives the company a foundation for AI-specific uses such as semantic caching and vector retrieval, without presenting Valkey as a general-purpose replacement for every database category.

The bottom line

Valkey Managed Database is best understood as the memory and retrieval layer in Akamai’s distributed-AI architecture. Its value proposition is straightforward: keep hot AI data in memory, place it nearer to inference and users, reuse results when possible, support vector-based retrieval, and offload database operations to a managed service.

The strongest use cases are latency-sensitive assistants, RAG pipelines, semantic caching, and stateful agents. The launch is strategically significant because it connects Akamai’s edge-inference ambitions to the data infrastructure those applications require. But the product is still described as limited availability, and claims about exact global coverage, channel-partner availability, or specific future locations require additional product or partner documentation before they can be treated as confirmed.