Everpure launches Data Stream for enterprise AI data

Everpure has launched Data Stream, aimed at companies preparing data for artificial intelligence.

The product is designed to work with data where it already sits, rather than requiring it to be copied into separate systems.

Everpure is positioning the launch around a problem many businesses face as they move AI projects from trials into day-to-day use. While market attention has centred on chips and models, the company pointed to research suggesting that data quality and fragmented information remain major obstacles.

A commissioned IDC survey cited by Everpure found that 94% of IT leaders see data quality as the deciding factor in AI success. It also found that 45% of businesses are seeking a unified data platform to bring together fragmented environments for AI work.

Data challenge

Data Stream is intended to help companies prepare unstructured data for uses such as natural language queries, search and analysis. It is designed to avoid moving or duplicating critical information, creating a path to use existing enterprise data in production AI systems.

The product forms part of a wider strategy that also includes Everpure Data Intelligence, previously known as 1touch, along with the company's FlashBlade and Portworx products. According to Everpure, Data Intelligence discovers, classifies and maps enterprise information across software-as-a-service, cloud, on-premises and mainframe environments, while applying access controls and governance policies.

That approach reflects a broader issue in corporate AI deployments. Organisations often hold data across separate systems and formats, making it harder to feed accurate information into AI models while maintaining security and compliance.

Everpure said Data Stream extends the NVIDIA AI Data Platform reference design. It described the system as a way to replace manual ingestion and manipulation with a pipeline that runs from ingestion to inference.

"We are undergoing a massive capital supercycle in AI, where the defining factor between industry icons and those who disappear is the ability to adapt," said Robert Lee, Chief Technology Officer, Everpure.

"The winning AI architecture requires a unified platform that allows businesses to start small with immediate use cases and seamlessly scale to exabyte capacity. Everpure solves this challenge by delivering a trusted, secure, and high-performance data pipeline that accelerates time-to-results for an enterprise's data," Lee said.

NVIDIA link

NVIDIA is also backing the architecture tie-up. Everpure said it is developing additional AI storage work with NVIDIA STX, which uses NVIDIA Vera and the NVIDIA BlueField-4 STX storage processor.

"Building the next generation of AI factories requires a data architecture that seamlessly bridges secure, governed enterprise data with accelerated computing," said Jason Hardy, Vice President of Storage Technology at NVIDIA.

"Everpure's integration with the NVIDIA AI Data Platform provides the infrastructure foundation organizations need to scale from AI experimentation to full-production intelligence," Hardy said.

Everpure also linked the launch to pressure on companies to keep expensive AI computing resources fully utilised. It argued that fragmented storage pipelines can leave graphics processors waiting for data during both training and inference.

The company said its FlashBlade systems are intended to address that bottleneck by delivering low latency and allowing storage and compute to scale separately. Customers can begin with FlashBlade//S and expand to FlashBlade//EXA, while Portworx manages containerised AI pipelines from the edge to the data centre.

One customer highlighted the economic impact of underused AI hardware. STN said it had standardised on FlashBlade to remove data bottlenecks in large AI workloads.

"Idle GPUs are economically destructive," said Sabur Mian, Chief Executive Officer and Founder of STN.

"We standardized on Everpure FlashBlade because it eliminates the data bottlenecks that typically stall massive AI workloads. FlashBlade//EXA allows us to scale thousands of GPUs with up to 800 Gbps of throughput per node while maintaining rock-solid latency, driving up to 20% performance improvement for our customers," Mian said.