Big-data management company Cloudera Inc. said today it’s partnering with storage infrastructure firm Vast Data Inc. on what the companies have termed a “unified AI factory.”

They claim to have designed and built a highly scalable production environment that can be used to continuously ingest, refine, govern and deliver vast volumes of enterprise data that can be used to train and run artificial intelligence systems. It’s a jointly developed infrastructure platform that can run both on-premises and inside cloud environments, giving enterprises the flexibility to deploy AI in the most suitable place based on their performance, cost and compliance needs.

The companies say it’s a timely innovation, because enterprises are scrambling to try and automate more business processes with autonomous AI agents. But as they do so, they’re quickly finding out that their existing data architectures simply aren’t up to the job, because they were never designed to handle the continuous data pipelines needed to funnel massive amounts of information to those AI systems.

As a result, many agentic initiatives have been held back by a problem known as “GPU starvation,” which is where the expensive graphics processing units that power those AI workloads are constantly sitting idle, waiting for the data they need to process. As a result, GPU utilization is much lower than it should be, which means enterprises struggle to pay for the cost of those expensive chips.

This is the challenge Cloudera and Vast Data think they know how to solve. They’re doing it by leveraging their respective strengths in data management and data storage to ensure companies have a way to continuously pump data into their AI systems with rapid throughput and low latency. In the case of Cloudera, it brings its data lakehouse architecture to the table, complete with portable and containerized data services that span data engineering, streaming, machine learning and governance.

Meanwhile, Vast Data is providing its Vast AI Operating System, which is based on a “disaggregated, shared everything” architecture. It’s designed to support exabyte-scale data volumes with unified high-performance storage and vector database services that are integrated with Nvidia Corp.’s cuVS, which is an open-source library that enables GPU-accelerated vector search and data clustering. By marrying Cloudera’s data engineering and AI services with Vast Data’s ultra-fast data platform, the companies believe they have everything they need to deliver highly optimized AI-ready data to GPU clusters with minimal latency.

Abhas Ricky, Cloudera’s chief business officer and general manager of applied AI, said the world’s largest enterprises have spent billions of dollars on GPUs, but cannot get anywhere close to full utilization of those resources. “This is due to data bottlenecks,” he said. “Our partnership with Vast Data eliminates GPU starvation and enables customers to build true AI factories, where data flows seamlessly from ingestion to insight.”

The collaboration doesn’t just fix the issue of starving GPUs. It also paves the way for autonomous AI agents to run in secure, private and sovereign AI environments that are governed by strict enterprise compliance controls too, the companies said.

For instance, enterprises can leverage Cloudera’s AI Inference service, accelerated with Nvidia’s NIM microservices, to deploy AI models and feed them with sensitive data. The companies say the new AI factory is available now through their respective sales teams and partner ecosystems, with more “industry-specific” variations of the platform set to arrive in future.

Vast Data co-founder Jeff Denworth said his company is partnering with Cloudera to provide the missing piece of the puzzle. Most enterprises already have all the data they need to feed AI, but they struggle to create the continuous pipelines for inference, fine-tuning and data analysis to take advantage of it, he said. “Cloudera and Vast are helping customers build AI factories that connect data, intelligence and infrastructure into a single operational platform for AI across hybrid environments,” he said.

Image: SiliconANGLE/Gemini