Microsoft unveils Azure HorizonDB, a Rust-powered PostgreSQL engine

Microsoft has announced Azure HorizonDB, a new cloud-native PostgreSQL database service built from the ground up with Rust, a memory-safe programming language. The service, currently in private preview, aims to deliver up to three times faster performance than standard PostgreSQL while dramatically reducing write latency and failover times.

The announcement positions Microsoft to compete more aggressively with Amazon Web Services’ Aurora database service, which has dominated the cloud-native relational database market. HorizonDB introduces several architectural innovations designed specifically for mission-critical workloads that require both high performance and reliability.

Rust Foundation Promises Memory Safety and Performance

The most significant technical decision behind HorizonDB is Microsoft’s choice to rewrite the storage engine entirely in Rust, departing from PostgreSQL’s traditional C-language foundation. Rust has gained traction in systems programming for its ability to prevent memory-related bugs and security vulnerabilities while delivering performance comparable to C.

Memory safety issues — buffer overflows, use-after-free errors, and similar bugs — have historically been the source of numerous database vulnerabilities. By using Rust, Microsoft aims to eliminate entire categories of security risks while maintaining the performance characteristics enterprises expect from database systems handling thousands of transactions per second.

The company has also rewritten several PostgreSQL extensions in Rust, extending the memory safety benefits beyond the core storage layer into commonly used database functionality.

Cloud-Native Architecture Reduces Latency

HorizonDB’s architecture separates compute and storage layers, a design pattern that has become standard for cloud-native databases. However, Microsoft claims its implementation offers unique advantages in write performance through innovations in how it handles the write-ahead log (WAL) — the mechanism databases use to ensure data isn’t lost if a system crashes.

Traditional PostgreSQL deployments write data multiple times during transaction processing, creating network overhead in distributed environments. HorizonDB’s architecture leverages Azure’s zone-redundant storage to enable single-round-trip commit latency across availability zones, reducing the network overhead that typically slows distributed database writes.

The system also implements sharding at the storage layer, distributing data across multiple storage nodes to improve read I/O efficiency and enable faster point-in-time recovery operations. This approach allows HorizonDB to scale storage independently of compute resources, automatically expanding up to 128TB as needed.

Impressive Performance Claims

Microsoft presented performance benchmarks at its recent Ignite conference showing significant improvements over standard PostgreSQL deployments:

Transactional workloads: HorizonDB delivers up to 3x faster throughput for online transaction processing (OLTP) workloads compared to self-hosted PostgreSQL.

Failover times: The service achieves failover in low single-digit seconds, significantly faster than traditional PostgreSQL high-availability configurations that can take 30 seconds or more.

Write latency: By reducing network round-trips and optimizing the WAL process, HorizonDB aims to provide consistently lower commit latencies than distributed PostgreSQL alternatives.

Vector search: Built-in DiskANN indexing for AI workloads reportedly delivers up to 3x faster vector similarity searches compared to standard PostgreSQL with pgvector extension.

However, these benchmarks have not yet been independently verified, and real-world performance will depend on specific workload characteristics and configurations.

Deep AI Integration

HorizonDB includes native integration with Azure’s AI services, reflecting Microsoft’s broader push to embed artificial intelligence capabilities across its product portfolio. The service features built-in DiskANN vector indexing, an algorithm developed by Microsoft Research that enables efficient similarity searches across high-dimensional data — a common requirement for AI applications.

The platform also offers AI model management capabilities, allowing developers to store, version, and deploy machine learning models alongside their data. This integration aims to simplify the architecture for AI-powered applications by eliminating the need for separate vector databases or model serving infrastructure.

For organizations building retrieval-augmented generation (RAG) systems or other AI applications that combine traditional database queries with vector similarity searches, HorizonDB’s integrated approach could reduce both complexity and latency compared to architectures that span multiple specialized databases.

Enterprise Security and Compliance

The service includes enterprise-grade security features expected for mission-critical workloads. HorizonDB supports Microsoft Entra ID (formerly Azure Active Directory) authentication, allowing organizations to manage database access through their existing identity systems rather than maintaining separate database credentials.

Azure Private Link support enables database connections over private network paths, keeping traffic off the public internet. This capability is particularly important for regulated industries with strict data sovereignty and network isolation requirements.

The zone-redundant storage architecture provides built-in resilience against datacenter failures, with data automatically replicated across multiple availability zones within a region. Combined with the fast failover times, this design aims to minimize both planned and unplanned downtime.

Competing With Amazon Aurora

Azure HorizonDB enters a market where Amazon Aurora has established itself as the leading cloud-native relational database. Aurora, launched in 2014, pioneered many of the architectural patterns HorizonDB now employs, including compute-storage separation and distributed, fault-tolerant storage systems.

Aurora supports up to 15 read replicas and offers impressive availability guarantees backed by Amazon’s infrastructure. The service has also claimed 3x performance improvements over standard PostgreSQL, setting a benchmark that HorizonDB now matches in Microsoft’s testing.

The competitive battle will likely focus on several dimensions: actual performance in production workloads, pricing (which Microsoft has not yet disclosed for HorizonDB), ease of migration from existing PostgreSQL databases, and integration with each cloud provider’s broader ecosystem of services.

Microsoft’s Rust-based approach could provide advantages in security and long-term maintainability, though these benefits may be less immediately visible than raw performance metrics. The integrated AI capabilities also differentiate HorizonDB for organizations building modern applications that combine traditional transactional data with vector embeddings and machine learning models.

Technical Specifications and Scale

HorizonDB supports clusters with up to 192 vCores distributed across primary and replica nodes, with as many as 15 read replicas to handle read-heavy workloads. Storage automatically scales to 128TB, eliminating the need for capacity planning and manual storage provisioning.

The service is currently available in private preview in four Azure regions: Central US, West US3, UK South, and Australia East. Microsoft has not announced a timeline for general availability or pricing, though enterprises interested in testing the service can apply for private preview access through Azure’s portal.

Migration Path and PostgreSQL Compatibility

One critical question for potential HorizonDB adopters is how easily existing PostgreSQL workloads can migrate to the new service. While Microsoft maintains PostgreSQL compatibility at the SQL and protocol level, the reimplemented storage engine and Rust-based extensions may introduce subtle behavioral differences from standard PostgreSQL.

Organizations with complex PostgreSQL deployments, particularly those using advanced features or extensions, will need to thoroughly test compatibility during the private preview period. Microsoft has not yet published detailed migration documentation or compatibility matrices, information that will be essential for enterprises evaluating HorizonDB.

The service’s cloud-native architecture may also require application changes for organizations currently running PostgreSQL with traditional high-availability configurations like streaming replication. Understanding these migration considerations will be important for realistic adoption planning.

What’s Missing: Pricing and SLAs

Microsoft has not disclosed pricing for Azure HorizonDB, making cost comparisons with Aurora or other alternatives impossible at this stage. Cloud database costs typically include charges for compute resources, storage capacity, I/O operations, and data transfer, with the specific pricing model significantly impacting total cost of ownership.

Similarly, formal service level agreements (SLAs) have not been published. While Microsoft claims low single-digit failover times and emphasizes availability through zone-redundant storage, the contractual commitments around uptime, performance, and support response times remain undefined.

These details will be crucial for enterprises making build-versus-buy decisions or choosing between cloud database services. Until pricing and SLAs are available, organizations can evaluate HorizonDB’s technical capabilities but cannot make complete financial or risk assessments.

Industry Context and Rust Adoption

Microsoft’s decision to build HorizonDB with Rust reflects broader industry trends toward memory-safe languages for systems programming. Google, Amazon, and other technology companies have been increasingly adopting Rust for infrastructure projects where security and reliability are paramount.

The Linux kernel community has begun accepting Rust code, and major infrastructure projects from web servers to network protocols are being rewritten in the language. HorizonDB represents one of the most significant production deployments of Rust in the database space, potentially validating the language for other database vendors considering similar transitions.

If HorizonDB successfully delivers on its performance and security promises, it could accelerate Rust adoption in the database industry, where C and C++ have dominated for decades despite their memory safety challenges.

What Comes Next

Organizations interested in Azure HorizonDB can apply for private preview access through Azure’s portal. Microsoft is likely gathering performance data, compatibility feedback, and usability insights from early adopters to refine the service before general availability.

The timeline for public launch remains unclear, though database services typically spend months in preview before becoming generally available. Enterprises with immediate PostgreSQL needs should plan accordingly, as HorizonDB may not be an option for new projects launching in the near term.

As more organizations test the service and share experiences, a clearer picture will emerge of HorizonDB’s real-world performance, migration challenges, and total cost of ownership. For now, Microsoft’s architectural innovations and performance claims have generated significant interest in the database community, setting high expectations for the service’s eventual public launch.

Follow us on Bluesky , LinkedIn , and X to Get Instant Updates