If the very definition of a supercomputer is rapidly evolving, its data infrastructure has no choice but to keep pace.
But the sheer breadth of what systems like NERSC’s upcoming Doudna supercomputer aim to accomplish has become a unique technical challenge.
From fusion reactors and dark-energy cosmology pipelines to genomics and quantum simulation frameworks, Doudna is expected to ingest, analyze, and interact with datasets in volumes and velocities that would have sounded like science fiction just a few years ago.
And because scientific workloads no longer wait patiently in batch queues, its storage must accommodate something utterly new: the messy immediacy of real-time science.
This sprawling diversity in workload profiles, a universe of complexity that ranges from streaming instrument data requiring sub-second responsiveness to massive AI models demanding continuous flash-driven throughput, is precisely the kind of challenge VAST Data was born to tackle.
But it’s one thing to talk in the abstract about “AI-native infrastructure” or “next-gen data architecture.” It’s quite another to deploy a system capable of dynamically reshaping itself around the needs of a national research community.
With the Doudna deployment at NERSC, VAST isn’t just showcasing the raw scale and flexibility of its DASE architecture, it’s rewriting expectations about how storage operates at the bleeding edge of computer science.
The new infrastructure behind Doudna needs to be not only fast but intelligently adaptive, with workload-aware quality-of-service rules, GPU-native data paths, and predictive caching mechanisms. In this machine, all are precisely calibrated to match whatever workloads the lab’s community will throw at it.
Take, for instance, the near-real-time data streaming required by astrophysics observatories or experimental fusion energy trials. These jobs can’t just pause for lengthy queries or delayed retrievals.
Meanwhile, machine-learning-heavy pipelines, from protein-folding simulations to quantum simulations and predictive cosmology models, need sustained GPU-friendly throughput at massive scale. No downtime, no bottlenecks, and certainly no room for traditional assumptions about hierarchical, tiered data handling.
The VAST deployment at NERSC has to function less as a classic “storage array” and more like a hyper-efficient operating system, one whose main task is intuitively matching data to compute resources in real time.
What’s particularly challenging is how Doudna’s storage infrastructure has to evolve in real time alongside NERSC’s ambitious user base.
To accommodate quantum simulation efforts, VAST’s architecture must support the seamless integration of quantum-inspired workloads that have yet to fully crystallize in terms of demands. At the same time, established simulation workloads (like high-energy physics or climate modeling) continue to surge in complexity, scale, and urgency, often shifting abruptly in their performance profiles and needs.
Doudna demands flexibility that is both predictive and instantly reactive, learning from the changing nature of the workloads themselves.
There’s elegance here. VAST isn’t just scaling for capacity or throughput, it’s architected for these new realities of science.
By using flash-native object storage built around fine-grained metadata, NVMe-oF connectivity, and software-defined QoS, VAST ensures that resources adapt fluidly to both expected and unforeseen data scenarios.
The infrastructure powering Doudna is a live demonstration of what it means to engineer not just for today’s workloads, but for the unknown territory of tomorrow’s science.
In taking on the Doudna supercomputer’s sprawling technical landscape, VAST is sending a clear signal. Data can no longer be relegated to a passive background role. It must actively anticipate, interpret, and respond to the evolving nature of scientific computing, transforming itself from a static repository into a dynamic, workload-intelligent fabric that sustains and propels innovation.
