This month VAST Data announced integrations with both Microsoft Azure and Google Cloud, significantly expanding its hyperconverged approach to AI Storage, what its calls its “AI Operating System,” further into major public cloud environments.

Azure customers will gain access to VAST’s complete data services suite running on Azure infrastructure, while Google Cloud users receive the first fully managed VAST AI OS service. Both integrations emphasize eliminating data migration barriers and supporting agentic AI workloads.

Technical Details

VAST Data’s cloud integrations introduce its “AI Operating System” to Azure and Google Cloud with distinct implementation approaches that reflect each provider’s infrastructure architecture and service model.

Azure Integration

The Azure sees VAST’s infrastructure running directly on Microsoft’s platform (rather than as a managed service). Enterprises using this integration will interact with VAST’s platform through Azure’s existing operational frameworks.

The implementation includes several technical components:

Core Infrastructure Foundation

VAST’s Azure presence leverages Microsoft’s Laos VM Series with Azure Boost Accelerated Networking. The platform implements VAST’s Disaggregated, Shared-Everything (DASE) architecture, which separates compute and storage scaling within Azure’s infrastructure model, allowing enterprises to expand compute or storage resources independently.

Data Platform Components

The integration delivers multiple data service layers that VAST claims operate as unified infrastructure:

• VAST DataStore provides multi-protocol support spanning NFS, SMB for file access, S3-compatible object storage, and block protocols, eliminating protocol translation overhead for diverse workloads.
• VAST DataBase combines what VAST describes as transactional performance with analytical query capabilities, positioning itself between traditional databases and data warehouses in the architectural stack.
• Similarity Reduction technology addresses storage efficiency through data deduplication and compression, though actual storage savings vary significantly based on data characteristics and workload patterns.

AI-Specific Capabilities

VAST emphasizes two components specifically targeting AI workloads:

• InsightEngine delivers stateless compute and database services to accelerate vector search, RAG pipelines, and data preparation tasks by executing these operations where data resides rather than requiring data movement
• AgentEngine orchestrates autonomous agents operating on real-time data streams, enabling continuous AI reasoning across distributed environments

VAST claims that these components keep Azure GPU and CPU clusters fully utilized through high-throughput data delivery, intelligent caching, and metadata-optimized I/O. However, sustained GPU saturation depends heavily on workload characteristics, data locality, and network performance, all factors that vary considerably across different AI model architectures and training scenarios.

Hybrid Connectivity Model

VAST’s DataSpace establishes a global namespace that the company claims reaches exabyte scale, creating unified data visibility across on-premises and Azure deployments.

Enterprise users can theoretically burst from local infrastructure to Azure for GPU-accelerated workloads without data migration or reconfiguration.

Google Cloud Integration

The Google Cloud implementation differs fundamentally from Azure by offering VAST AI OS as a fully managed service available through Google Cloud Marketplace. This managed service model reduces operational overhead for customers but introduces dependencies on Google Cloud’s service management framework.

Managed Service Deployment

Google Cloud customers can deploy VAST AI OS through marketplace provisioning, which VAST describes as enabling production readiness “in minutes.”

The managed service approach allows infrastructure provisioning, monitoring, and maintenance to be handled through Google Cloud’s operational framework.

DataSpace Global Connectivity

As part of the announcement, VAST demonstrated DataSpace capabilities connecting clusters separated by over 10,000 kilometers (linking US and Japanese locations) while maintaining what the company characterized as “near real-time” data access.

The demonstration ran inference workloads using vLLM across this distributed configuration, enabling workload placement on TPUs in one region and GPUs in another without data replication.

This capability addresses the genuine enterprise challenge of running AI workloads where computational resources are available or cost-effective while avoiding months-long data migration projects.

However, “near real-time” performance claims require scrutiny, as network latency across intercontinental distances introduces unavoidable physical constraints that affect certain workload types more severely than others.

TPU Performance

VAST published performance results for Google Cloud TPU virtual machine integration using Meta’s Llama-3.1-8B-Instruct model. The results show that model load speeds are “comparable to some of the best options available in the cloud” with predictable cold-start performance.

While these results suggest VAST AI OS can effectively support TPU-based training and inference, several critical details remain unspecified:

• Exact model load times in quantified metrics rather than comparative statements
• Network configuration and bandwidth between VAST storage and TPU instances
• Performance variability across different model sizes and architectures
• Cost comparison between VAST AI OS and native Google Cloud storage options

The absence of these sorts of benchmarks limits objective assessment of these performance claims against alternative architectures.

Technical Capabilities for Google Cloud Deployments

Organizations deploying VAST on Google Cloud gain access to several operational capabilities that address hybrid AI infrastructure challenges:

• Instant Data Availability: DataSpace presents existing on-premises datasets to Google Cloud workloads through a consistent global namespace, using intelligent streaming to transfer only required data subsets rather than full dataset replication
• Selective Data Placement: Organizations retain control over migration, replication, and caching decisions while maintaining unified namespace visibility and governance policies across all locations
• Unified Access Controls: VAST applies consistent access controls, audit logging, and retention policies across on-premises and Google Cloud environments, simplifying compliance management for distributed deployments
• Multi-Protocol Support: DataStore and DataBase components provide NFS, S3, and database access patterns from a single platform, reducing pipeline complexity for workflows spanning data preparation, training, inference, and analytics

Architectural Commonalities Across Cloud Integrations

Both Azure and Google Cloud integrations share several fundamental architectural elements that define VAST’s hybrid cloud strategy:

Unified Data Fabric Approach

VAST’s DataSpace is the connective tissue spanning disparate infrastructure locations, eliminating what the company characterizes as “data silos” that fragment enterprise AI initiatives. This unified namespace model addresses legitimate operational challenge, such as the complexity and cost of maintaining data consistency across multiple storage systems with different access patterns and governance frameworks.

Performance Claims and Validation Requirements

In its announcement, VAST emphasizes keeping GPU and TPU accelerators “saturated” with data and maintaining “predictable performance from pilot to multi-region scale.” These claims address critical pain points in AI infrastructure, such as the challenge of ensuring expensive accelerator resources remain productive rather than idle while waiting for data.

Governance and Compliance

Both integrations emphasize maintaining consistent governance, security, and compliance policies across hybrid deployments. VAST claims to provide unified access controls, audit capabilities, and retention policy management regardless of data location. For enterprises operating under regulatory frameworks such as GDPR, HIPAA, or financial services regulations, this unified governance model offers potential operational simplification.

Analysis

VAST Data’s parallel integrations with Microsoft Azure and Google Cloud are significant strategic moves that position the company’s hyperconverged AI solution, its “AI Operating System,” as infrastructure for hybrid enterprise AI deployments.

Data fragmentation across hybrid environments, migration complexity, and the need for unified governance are very real problems for enterprises operating AI in the cloud. VAST addresses these challenges with architectural approaches that differentiate it from both traditional enterprise storage vendors and cloud-native alternatives.

VAST’s multi-cloud strategy show strong alignment with emerging enterprise requirements for AI infrastructure that maintains flexibility across cloud providers while supporting demanding performance requirements.

The company’s emphasis on eliminating data migration barriers particularly resonates for enterprises with substantial existing data estates seeking to leverage cloud computational resources without months-long migration projects.

For enterprises committed to hybrid AI architectures and willing to invest in comprehensive platform evaluation, VAST’s Azure and Google Cloud integrations merit serious consideration as potential foundations for distributed AI infrastructure. The partnerships show VAST continuing to gain momentum as it establishes itself as a credible alternative to both legacy enterprise storage platforms and cloud-native services.

Competitive Outlook & Advice to IT Buyers

These sections are only available to NAND Research clients and IT Advisory Members. Please reach out to [email protected] to learn more.