VAST & Nvidia unveil integrated GPU-first AI data stack

VAST Data has expanded its collaboration with Nvidia, launching an end-to-end AI data stack that runs the VAST AI Operating System directly on GPU-accelerated servers.

The design integrates Nvidia software libraries into VAST services for data ingestion, retrieval, analytics, and inference. VAST says the approach reduces data bottlenecks and simplifies deployments that often depend on separate storage, database, and AI infrastructure products.

VAST announced the update at its Forward conference in the US, pointing to demand from organisations that have prioritised AI but still lack integrated systems for production workloads.

Running on GPUs

At the centre of the update is the VAST AI OS running directly on Nvidia-powered servers. VAST says this makes GPU servers part of the core platform, rather than an external compute layer attached to storage.

VAST also introduced a new server line, CNode-X, described as a new generation of Nvidia-Certified Systems. It is intended for Nvidia GPU-accelerated clusters and for hosting VAST software services on the same infrastructure.

VAST says the shift allows it to orchestrate AI pipelines alongside analytics and search within a single software stack, reducing the operational overhead of integrating multiple platforms for continuous AI workloads such as retrieval augmented generation and agent-based systems.

Product components

CNode-X servers provide the computing foundation for the VAST AI OS, with Nvidia libraries and APIs embedded in core services including VAST DataEngine and VAST DataBase.

In the database layer, VAST is adding GPU-native acceleration for SQL analytics. The system combines storage-side techniques such as intelligent data layout, pushdown, and filtering with GPU-accelerated query execution.

For GPU-accelerated SQL execution, VAST uses Sirius, an open-source query engine based on Nvidia cuDF. Early benchmarking of Sirius showed up to a 44% reduction in query time and up to an 80% reduction in query cost, according to the company.

For vector search and retrieval, VAST is embedding the Nvidia cuVS library to accelerate vector search and data clustering. VAST says this can reduce retrieval latency for use cases that rely on real-time context, including production retrieval augmented generation pipelines.

In the data pipeline layer, VAST plans to deploy and support Nvidia NIM microservices across CNode-X. It is also open-sourcing production-ready VAST DataEngine blueprints for AI pipelines targeting video intelligence, enterprise document retrieval augmented generation, and genomics research.

VAST also supports the Nvidia Context Memory Storage platform, including configurations with Nvidia BlueField-4 DPUs and Spectrum-X Ethernet networking. The goal is faster access to shared key-value cache and lower time-to-first-token for long-context, multi-agent inference.

Partner supply

VAST plans to bring CNode-X servers to market through OEM partners including Cisco and Supermicro, giving customers a choice of hardware suppliers while keeping the VAST software and support model consistent.

VAST says certified OEM configurations offer a more predictable route to production deployments. It also argues that as AI pipelines become continuous systems, infrastructure must keep GPUs utilised across retrieval, analytics, and vector search workloads.

Renen Hallak, founder and CEO of VAST Data, said the announcement reflects the company's long-term product direction.

"Ten years ago, we set out to build a system that could continuously refine data into intelligence and action," said Renen Hallak. "That future is here. By accelerating both compute and the data paths inside the VAST AI OS with NVIDIA, we're giving customers a faster, simpler way to operationalise retrieval, analytics, and agentic workflows as one coherent pipeline so AI can move from pilot to durable, production systems."

Nvidia CEO Jensen Huang framed the news as part of a broader shift in AI infrastructure.

"NVIDIA is reinventing every pillar of computing for AI. With VAST Data, we're transforming the storage of AI infrastructure," said Jensen Huang. "CNode-X is CUDA-accelerated at every layer to give AI agents persistent memory so they can work on complex problems over days or weeks, and eventually years, without forgetting - opening the world to the next frontier of AI."

Cisco and Supermicro also cited demand for integrated infrastructure in enterprise environments.

"AI doesn't scale on isolated components. It scales through integrated systems," said Jeremy Foster, SVP and General Manager, Cisco Compute. "Customers need infrastructure that keeps data secure and tightly aligned with intelligent networking and GPU-accelerated compute for an efficient, production-ready platform. Cisco's collaboration with partners like VAST and NVIDIA is delivering the enterprise-ready foundation organisations need to help securely scale AI with performance, resilience, and control."

"Production AI demands a new level of integration across compute, acceleration, and the data platform," said Charles Liang. "Together with VAST Data and NVIDIA, we're delivering a truly integrated AI Data Platform that removes complexity from enterprise AI. By bringing high-performance compute, scalable data infrastructure, and intelligent software together as one solution, we're enabling organisations to move from experimentation to production faster and unlock real business value from AI."