Introducing BoulderKV: A Global, Read-Optimized Key-Value Store

We’re excited to introduce BoulderKV, a global, read-optimized key-value store built on object storage.

The Problem

Modern applications need to serve data globally with low latency, but traditional databases can be expensive and complex to scale worldwide. Meanwhile, many workloads don’t need strong consistency—they need durability, availability, and predictable costs at scale.

Our Solution

BoulderKV is designed specifically for workloads where:

• Reads dominate writes
• Datasets are large, but only a small fraction is frequently accessed
• Eventual consistency is acceptable
• Global distribution matters

By building on object storage (like S3), BoulderKV delivers:

• Low-latency reads through intelligent caching
• High durability from battle-tested object storage
• Global scale without operational complexity
• Predictable costs that scale with usage

Perfect Use Cases

BoulderKV shines for:

• ML Feature Stores: Serve training and inference features globally
• Inference Caches: Store model predictions at scale
• Historical Data: Order history, game state, user profiles
• Configuration Data: Feature flags and app configuration
• CDN-like Workloads: Distribute read-heavy content worldwide

Architecture Philosophy

We favor simplicity over complexity. BoulderKV uses asynchronous replication to scale globally while keeping the system easy to understand and operate. Updates propagate eventually, which allows us to maintain high availability and low costs.

What’s Next

We’re building BoulderKV in the open and would love your feedback. Whether you’re running ML infrastructure, building global applications, or managing large-scale data, we’d love to hear about your use cases.

Stay tuned for more technical deep dives on our architecture, performance benchmarks, and best practices.