The main constraint in enterprise AI is no longer compute. It is data.
For most organisations, the challenge lies in moving and managing large datasets across environments. Shifting petabytes of data from on-premises systems to cloud GPUs is slow, expensive, and operationally complex. This issue, known as data gravity, often limits how quickly companies can train or deploy AI models.
VAST Data and Google Cloud just announced a partnership aimed at addressing that constraint. The companies are offering a fully managed service for VAST’s “AI Operating System” on Google Cloud, designed to let enterprises run AI workloads across on-premises and cloud environments without moving most of their data. The service is available via the Google Cloud Marketplace and combines VAST’s data-management layer with Google’s compute infrastructure.
Integrating On-Prem Data With Cloud Compute
The joint offering uses VAST’s DataSpace technology, which creates a unified namespace linking on-premises and cloud clusters. In one demonstration, clusters located more than 10,000 kilometres apart (in the United States and Japan) accessed the same dataset in near-real-time.
According to VAST, tests using Meta’s Llama 3.1-8B model recorded model-load speeds comparable to local NVMe storage, even during cold-starts.
By streaming only the required portions of data instead of replicating entire datasets, VAST says the system aims to reduce bandwidth costs and management overhead. It also claims that deployments that previously took months can be initiated within minutes via provisioning on Google Cloud. VAST Data vice-president Aaron Chaisson described the approach:
“Moving large datasets is one part bandwidth and nine parts copy management, permissions drift, backup jobs, audit gaps, tool sprawl, etc. With VAST AI OS, it’s one system across sites: one namespace, one permission structure, one audit trail, one point of control, one source of truth.”
For Google Cloud, the integration extends its hybrid-AI offering: it gives customers who retain large on-prem datasets (often for compliance or latency reasons) a managed way to connect those sources with cloud compute services.
Turning Storage Architecture Into AI Infrastructure
VAST Data was founded in 2016 by Renen Hallak, Shachar Fienblit, Jeff Denworth and Alon Horev. The company’s early products were based on a “Disaggregated and Shared Everything” (DASE) architecture, which separates compute and storage resources to improve scalability.
Its initial goal was to replace mechanical media with flash-based systems and remove the need for storage-tiering: what VAST once described as “an extinction-level event for mechanical media in the enterprise data center.”
The company now positions itself around what it calls an AI Operating System, combining storage, database and compute orchestration under one software platform. By early 2025, VAST reported approximately US $200 million in Annual Recurring Revenue and said it was free-cash-flow positive.
A funding round in December 2023 raised US $118 million, valuing the company at US $9.1 billion; fundraising discussions reported in mid-2025 suggested valuations approaching US $30 billion.
The Google Cloud agreement follows a series of partnerships that have expanded VAST’s role in AI infrastructure. In early November 2025, VAST signed a US $1.17 billion deal with CoreWeave to supply its data platform for that company’s GPU-cloud service.
Days later, it was selected by Nscale as the data layer for an AI platform expected to scale to 300,000 NVIDIA Grace Blackwell GPUs.
Customers such as Pixar Animation Studios describe using VAST’s systems to consolidate large data archives for potential use in AI workloads.
“VAST is allowing us to put all of our rendered assets on one tierless cluster of storage… which offers us the ability to use these petabytes of data as training data for future AI applications.,” says Eric Bermender, head of data-centre infrastructure at Pixar.
For enterprises working on AI, the VAST Data-Google Cloud service presents a practical method to connect large, immovable data stores with cloud compute resources. Organisations can maintain their existing storage footprints, typically driven by cost, compliance or latency, while accessing elastic compute via Google’s infrastructure.
Analysis from Gartner highlights the challenge: “The success of ML and AI initiatives relies on orchestrating effective data pipelines that provision the high quality of data in the right formats in a timely manner during the different stages of the AI pipeline.”
By focusing on unified access and real-time data availability, the partnership positions VAST Data as a provider of infrastructure that enables compute and data to operate in tandem with fewer constraints. Data gravity is not eliminated, rather it is addressed through architecture.
