Enterprises don’t struggle to buy AI tools. They struggle to build production-ready systems.
Today, the default approach toward assembling production AI systems requires stitching together any number of different tools at each layer of the stack, from GPU compute to agentic orchestration. And although every component at each layer might be capable of performing its specific job, they’re often not optimized to work together as a system. Many were designed for a previous generation of web-based workloads that demand much less of their data infrastructure than does artificial intelligence.
When assembled, the result is less of a cohesive system and more of a collection of parts. An AI application architecture that’s fragile, complex, and difficult to scale. We built the VAST AI OS as a counter to this status quo: It’s a tightly integrated system purpose-built for the most demanding data-centric workloads, including AI.
This week at our VAST FWD user conference, we announced VAST AI OS 5.5, as well as a collection of additional features and advancements that take the AI OS to the next level around some critical facets for running AI in production, including performance, governance, security, orchestration, and ecosystem coordination.
VAST AI OS 5.5
Highlights of the latest release include the new VAST Hyperscale Vector Index, delivering trillion-vector search with dramatically lower cost and latency; expanded native SQL analytics for running complex queries directly on the platform; integrated Kubernetes compute for orchestrating AI pipelines alongside the data; and S3 over RDMA for high-performance object storage. Together, these innovations simplify infrastructure, accelerate AI workflows, and provide a production-ready foundation for hyperscale, real-time intelligence.
GPU-accelerated performance
GPUs are firmly established as the default compute option for training AI models and running AI inference, but they can do so much more. With the VAST CNode-X server, we’re adding GPUs directly into VAST-certified hardware platforms and integrating the VAST AI OS with key libraries for GPU-accelerated SQL, vector search, and the management of AI models as microservices. As a result, customers will be able to collapse their AI inference and data-processing architectures while significantly improving performance on important workloads.
Governance for self-learning AI agents
For many large enterprises and organizations, systemic governance is a requirement applying AI to their critical applications. Trust is not a bolt-on feature; it must mediate every interaction. The introduction of the VAST PolicyEngine and the VAST TuningEngine extends the VAST AI OS into a trusted, auditable, and self-learning runtime for AI agents. PolicyEngine mediates every interaction between agents, tools, memory, and data according to fine-grained policies. TuningEngine learns from these interactions and policies, and updates the underlying models to better reflect customer use cases, rules, and real-time context.
Governance and improvement are no longer external processes, they are part of the operating system itself.
Always-on security intelligence
As AI systems scale, risk concentrates where data is accessed and transformed and where workloads run. Building on VAST’s native security and governance controls, new CrowdStrike integrations add continuous threat detection, workflow-level scanning, and automated response — paired with shared telemetry — to help customers identify and contain threats with minimal disruption.
Orchestrating AI across environments
AI pipelines no longer live in a single cluster or location. Training, inference, and data collection span public cloud, private datacenters, and sovereign environments. VAST Polaris introduces a global control plane that provisions, operates, and orchestrates VAST environments across these domains as a unified system. Infrastructure location becomes abstracted from operational control.
But infrastructure maturity alone is not enough. AI systems are ultimately judged by the workloads they can sustain, and multimodal video intelligence is one of the most demanding — requiring massive datasets, complex embeddings, and strict governance. By enabling customer-managed deployments of TwelveLabs’ models directly on the VAST AI Operating System, we are demonstrating that the architecture can support real, production-scale AI systems in controlled environments.
Building an operational ecosystem
AI at scale inherently involves working with multiple vendors. Compute, hardware, and networking providers all play a role, as do ISVs, cloud platforms, VARs and solutions integrators, and the broader builder community. Without coordination, integration risk increases and delivery slows. The VAST Cosmos Partner Program formalizes the ecosystem layer around the VAST AI Operating System, aligning validated architectures, integration standards, and repeatable deployment models. It provides the coordination framework required to turn platform capability into operational outcomes.
Bridging the gap from project to production
Taken together, these advancements reflect VAST’s deliberate architectural direction. Rather than building isolated features and check-the-box integrations, we’re building a unified AI operating system. This week’s announcements, paired with the amazing collections of customers and partners who came together at VAST FWD, underscores how the VAST AI Operating System can operate as a center of gravity for AI operations. We’re simplifying operations and securing data without sacrificing performance or scalability.
