As we navigate the complex landscape of 2026, the intersection of massive data estates and artificial intelligence has become the primary battlefield for enterprise success. I’m joined today by Maryanne Baines, a leading authority in cloud technology, to dissect the shifting paradigms of data management. We’ll be exploring how the transition from simple hardware to full-stack innovation is redefining the “AI factory,” addressing the critical bottlenecks that have long hindered GPU performance, and examining the new frontier of data sovereignty. Our conversation covers the move toward unified storage consoles, the introduction of zetta-scale file systems, and the urgent need to illuminate “dark data” to fuel agentic AI.
Since a vast majority of enterprise data is currently distributed across fragmented silos, how can organizations effectively overhaul their data estates to meet the rigorous demands of modern AI?
The reality facing most enterprises today is quite stark, as roughly 85% of data remains trapped in isolated silos across fragmented environments. This “dark data”—which includes everything from decades-old medical research to engineering documentation and operational files—is a literal goldmine that most companies simply cannot tap into. To move forward, organizations have to shift their focus from just collecting data to establishing a modern data architecture that prioritizes visibility and flow. We are seeing that 93% of organizations report at least one major data-related challenge blocking their AI ambitions, and that number only grows as they try to scale. By consolidating these disparate estates into a unified platform, businesses can finally provide the “memory and context” that AI agents need to be truly effective and autonomous.
With the launch of the AI Data Engine, we’re seeing a push toward cleaning up these data estates, but how does this technology actually help a researcher or a developer find value in a mountain of old files?
The AI Data Engine, or AIDE, acts as the essential layer that understands exactly what sits within a data estate, effectively turning raw storage into actionable memory. For someone like a medical researcher who might be sitting on decades of various assets and data sources, pulling that information together manually is a monumental, almost impossible task. AIDE streamlines this by offering heterogeneous data discovery, which allows for total visibility across any environment regardless of the underlying storage architecture. To encourage this transition, there is even a push to allow customers to map their data at no cost for six months, which helps them activate their data wherever it lives. It’s about moving beyond simple storage and into a world where the system knows the context of the information it holds, making it “AI-ready” from day one.
We’ve heard a lot about the “AI factory” concept and the massive performance requirements it demands, so what specifically makes a system like NetApp Novus a game-changer for these high-performance environments?
NetApp Novus is being described as one of the most significant leaps in storage history because it directly tackles the recurring bottlenecks that have plagued GPU-hungry environments. This is a zetta-scale file system that boasts a staggering throughput of up to 100Tbps, which is essentially equivalent to unifying the entire world’s data center capacity that shipped just last year. Traditional storage architectures weren’t built for the dual pressure of billions of small reads during data loading and terabyte-scale checkpoint writes happening simultaneously. Novus changes the game by supporting up to 50,000 GPUs and ensuring that these expensive chips aren’t left idling while waiting for data. When you consider that a stalled read in a large cluster can lead to tens of millions of dollars in wasted capacity every single day, the importance of this architectural shift becomes very clear.
Given that power availability and cooling are now major constraints for growth, how can high-performance storage actually contribute to a more sustainable or energy-efficient AI operation?
It is a common misconception that high performance always equals higher power consumption, but the latest innovations prove otherwise. Many AI factories are currently seeing GPU utilization as low as 30% in certain instances, which represents a massive waste of both electricity and capital. By maintaining vital data flows and ensuring GPUs are constantly fed, the Novus architecture can actually reduce power requirements by up to 31% compared to other NFS storage alternatives. This is a significant saving that scales rapidly as an organization grows from a 10,000-GPU setup to a 50,000-GPU superintelligence environment. Efficient storage ensures that we aren’t just throwing more power at the problem, but rather making sure every watt consumed is actually contributing to the training or inference process.
As organizations expand into hybrid cloud setups with infrastructure in multiple locations, how do they maintain a consistent security posture without overwhelming their administrative teams?
The complexity of modern hybrid clouds—often involving four different teams across four different consoles—frequently leaves administrators unsure if their security policies are being enforced everywhere at once. The solution lies in moving toward a single control plane, like the NetApp Console, which provides a unified view of disparate storage infrastructure. Security in this age has to mean more than just an “identity check at the door”; it requires a new definition of zero trust that is built directly into the storage platform itself. This includes proactive protection and specialized tools for defenders to close the loop when ransomware strikes, because as we often say, detection without a clear path to recovery is just an expensive alarm. By integrating large language models directly into the management console, admins can now use natural language to monitor these complex environments, making high-level security more accessible.
With the rise of Keystone Sovereign, how is the industry addressing the growing need for data sovereignty and local regulatory compliance without sacrificing the flexibility of storage-as-a-service?
Data sovereignty has rapidly become a top-tier concern for enterprises, particularly those operating across the Atlantic where regulations are tightening. The Keystone Sovereign model addresses this by allowing customers to select dedicated sovereign options, ensuring that their hardware is hosted in specific regions like France or Germany. Beyond just location, it involves a commitment to using local staff for operational support, which is a critical requirement for many government and highly regulated industry contracts. This “sovereignty-focused” add-on allows businesses to enjoy the financial flexibility of a storage-as-a-service (STaaS) model while still meeting the strict legal requirements of their specific jurisdiction. It’s a way of proving that you don’t have to trade off innovation and cloud-like agility for the sake of compliance.
What is your forecast for the evolution of AI storage architectures over the next few years?
I expect that we will see a complete departure from traditional appliance-based deployment models as the “tail wagging the dog” dynamic—where hardware is beholden to limited software—finally disappears. We are moving toward a future where metadata scales independently from data via disaggregated architectures like Lattice, allowing for nearly infinite flexibility regardless of whether a company is training a frontier model or running inference at scale. Storage will no longer be viewed as a passive repository but as a high-speed, intelligent engine that is deeply integrated with the GPU clusters it serves. Within the next few cycles, the standard for success will be defined by “zettabyte-scale” capabilities, where the world’s total data output is processed through these highly efficient, sovereign, and self-healing storage fabrics. This evolution will turn the current data bottlenecks into a seamless highway, finally allowing the true potential of agentic AI to be realized across every global industry.
