Few people have had a better vantage point on the evolution of data infrastructure than Sven Breuner.
As the creator of the BeeGFS parallel file system and now as Field CTO at VAST Data, his career spans an era in which storage evolved from a specialized component of high performance computing into a foundational platform for AI.
On the latest episode of the Shared Everything podcast, Breuner reflects on that journey and explains why the industry’s priorities have shifted. He explains how where infrastructure was once judged almost entirely by raw performance, today’s challenge is helping people get more value from the data they already have.
When Breuner began developing BeeGFS at Germany’s Fraunhofer Center for High Performance Computing in 2005, the focus was straightforward. Parallel file systems competed on throughput, scalability, metadata performance, and reliability. Features such as distributed metadata were introduced to remove bottlenecks and improve efficiency, while usability and deployment became important differentiators from other file systems of the day.
Looking back, Breuner notes that the HPC community largely viewed storage through the lens of performance, assuming that faster systems naturally led to better outcomes. Over time, however, he came to recognize that even exceptional performance alone doesn’t guarantee better science or more productive research. The more important question became what users could actually accomplish with their data.
That realization mirrors a much broader transformation happening across the industry. Storage is increasingly expected to do far more than reliably hold files. Modern data infrastructure must support multiple access methods, integrate databases and indexing technologies, and provide the foundation for AI-driven discovery.
Breuner describes VAST’s long-term vision as building toward a “thinking machine” capable of helping organizations ask questions of their data in natural language while automatically finding or generating the information they need. While that vision remains aspirational, it reflects a growing expectation that infrastructure should actively help people understand and use information rather than simply store it.
Artificial intelligence is accelerating this shift because it changes how organizations interact with data he says. Traditional HPC simulations typically read complete datasets, perform calculations, and generate new outputs. AI and analytics workloads operate differently. They depend on searchable, indexed information that can be queried quickly to identify relevant patterns and relationships.
All of this, of course, requires a fundamentally different approach to infrastructure, one designed not only for moving data efficiently but also for making information discoverable and immediately useful. As he explains, AI excels at recognizing patterns, but it depends on well-organized data to do so effectively.
For research organizations and enterprises alike, adapting to AI is not a one-time upgrade but an ongoing process. Breuner argues that the organizations making the greatest progress are those willing to rethink long-held assumptions about their infrastructure rather than simply adding GPUs to existing environments.
Throughout the conversation, he also describes how capabilities such as workflow automation are becoming increasingly valuable. Instead of treating storage as a passive destination for files, modern platforms can automatically trigger downstream analysis, launch new workflows, transform datasets, or notify users when data is ready.
These kinds of data-centric capabilities allow infrastructure to become an active participant in the research process rather than simply supporting it.
Looking ahead, Breuner sees AI becoming an indispensable research assistant rather than a replacement for human expertise. Large language models and other AI systems will continue to improve at gathering information, summarizing knowledge, recognizing patterns, and accelerating routine tasks. Creativity, curiosity, and asking the right questions, however, remain uniquely human strengths.
The future of data infrastructure, he argues, is not simply about building faster systems. It is about creating platforms that help people and AI work together more effectively, turning ever-growing volumes of information into meaningful insight and ultimately enabling the next generation of scientific discovery.
