Data storage company Vast Data Inc. is putting the infrastructure in place for enterprises that want to use data retrieval to enhance the capabilities of their artificial intelligence systems.

Today it announced the launch of Vast InsightEngine with Nvidia, set to enter general availability early next year. It will allow organizations to ingest, process and retrieve all of their enterprise data, including files, objects, tables and data streams, securely and in real time, the company said.

The announcement came as Vast Data revealed a new initiative called Cosmos, which aims to transform how enterprises build and advance AI. It’s a community initiative focused on collaboration, with the goal being to create a comprehensive and supportive environment for innovation and growth. Its founding members include an array of AI practitioners, researchers, service providers and solutions integrators, all working together to democratize AI, the company explained.

Vast Data is best known for its storage management platform, which is used to manage and scale storage infrastructure across on-premises and cloud environments. With the growing interest in AI, it has turned its attention toward building a comprehensive data platform for AI applications with advanced processing capabilities.

The company believes there’s a big opportunity for this, because organizations need an easy way to handle the vast amounts of information needed for training and fine-tuning AI models. Using Vast’s platform, it’s easier to manage and use complex, unstructured data types such as videos and images in AI applications.

A platform for real-time RAG

With today’s announcement, the company is turning its attention to retrieval-augmented generation, or RAG — which Vast prefers to call simply retrieval. It’s a technique that businesses can use to enable large language models to tap into their own, proprietary datasets to expand their knowledge. An example of this might be an AI-powered customer service chatbot that needs to access the company’s internal knowledge base to respond to customer queries, for instance.

Steve McDowell of NAND Research Inc. told SiliconANGLE that RAG has become essential for enterprises to bring their data to pretrained foundation models, but implementing it is a major challenge for developers.

For RAG to work properly, organizations need a storage infrastructure that’s able to classify and search both structured and unstructured data and also support semantic approaches such as vector and knowledge graphs. This is what Vast Data is doing with Vast InsightEngine, which was built in collaboration with Nvidia Corp. and provides a unified system for delivering real-time information to AI systems.

Notably, Vast InsightEngine integrates with Nvidia’s NIM Microservices, which provides the advantage of embedding the semantic meaning of incoming data. These embeddings are stored in the Vast DataBase within milliseconds, so as to accelerate time to insights and simplify data pipeline operations, the company said.

Vast InsightEngine makes use of the company’s DataEngine to trigger the Nvidia NIM embedding agent the moment any new data comes into the system. This facilitates the creation of vector embeddings and graph relationships from unstructured data in real time, eliminating the traditional delays associated with batch processing. As a result, companies will benefit from “near-instant” availability of fresh insights that can be delivered straight to their AI models.

Other components include Vast’s highly scalable Vast DataBase and a unified data architecture that integrates storage, processing and retrieval of all data types into a single platform, with data indexing performed at the source of the data. In doing this, Vast said, it does away with the need for maintaining separate data lakes and external software-as-a-service platforms in order to reduce costs and complexity.

In a press briefing, Vast co-founder and Chief Executive Renen Hallak said Vast InsightEngine is both the world’s first real-time data retrieval system and the world’s first hyperscale vector database. It’s essentially an operating system for AI, he added.

“We’re becoming the middle of the sandwich,” he explained. “We’re no longer just a storage system sitting next to a supercomputer. We’re part of the system, at the center of this new world.”

Key differentiators for Vast

McDowell said the main advantage of Vast InsightEngine is that it performs the entire RAG or retrieval process in a single platform, without moving data out of the storage platform to a vector database, then writing it back onto the storage.

“Instead of bouncing data to an external tool and incurring that performance penalty, Vast allows you to do that completely within their data platform,” the analyst said. “This reduces complexity while increasing performance and is a big win that gives Vast a huge competitive differentiator. Nobody else is doing this right now.”

A second key differentiator is the integration with Nvidia’s NIM services, which run natively within Vast’s platform, the analyst said. “This removes the other point where data exits the storage system, and will make data scientists very happy with the performance gains,” he added.

Not an easy sell

Despite having these two, significant advantages, McDowell said Vast will still have a lot of difficulty in selling its solution for RAG workloads, because adopting its platform could cause companies to be locked into it.

“Storage has always been replaceable — if you don’t like your NetApp array, you can always swap it out for a similar product from Dell or Pure Storage,” McDowell explained. “But by building dependencies between its storage layer and RAG or some other computation, Vast is really creating lock-in for customers. It’s not necessarily the wrong approach, but I doubt that every enterprise customer will want to create that dependency.”

For those that are considering Vast’s platform, the biggest question will be whether or not they want to retain the flexibility of legacy storage or go for the performance and scalability of a data platform that’s much more tightly integrated. But McDowell said it’s too early to know how many enterprises will choose the latter option.

“The new features and NIM integration are perfectly aligned to where the industry is in its broader AI journey,” the analyst added. “The solution is compelling, and Vast will not lose because of technology.”

Accelerating AI innovation

As for the Cosmos initiative, it’s all about helping Vast Data’s customers accelerate adoption of AI. To that end, Vast is building a comprehensive and interconnected ecosystem to facilitate conversations, share use cases and provide learning opportunities for every organization through labs, showcases and general research.

Through Cosmos, Vast’s customers will be able to explore a range of AI solutions and implement them more easily while participating in the wave of AI innovation, the company said. Of course, there will be a big focus on AI applications and systems that leverage Vast’s underlying storage infrastructure. The initiative, while fostering collaboration and encoraging accelerating AI adoption, is also designed to help to promote Vast as the storage platform of choice for AI workloads.

Founding members of the Cosmos community include Nvidia, Elon Musk’s xAI Corp., Super Micro Computing Inc., Deloitte Touche Ltd., World Wide Technology Inc., Cisco Systems Ltd., RunAI Inc., Dremio Inc. and CoreWeave Inc.

“With the Vast Data Platform at the center of this comprehensive, interconnected AI ecosystem of technology leaders and AI practitioners, Cosmos will help accelerate discovery, empowering innovation, and enabling the transformation of entire industries,” Hallak said.

Last month, Hallak sat down for an interview with John Furrier, executive analyst at theCUBE Research, to talk about his  company’s growing role in the AI ecosystem:

With reporting from Robert Hof

Photo: SiliconANGLE