AI storage strategy is taking on a larger role as enterprises move from successful experiments to systems that must deliver dependable business results. How organizations store, serve and manage data increasingly shapes the performance, cost and practicality of artificial intelligence deployments.
The challenge extends beyond choosing faster hardware. Evolving workloads require enterprises to rethink how their infrastructure supports data-intensive applications. Turning that infrastructure into a production foundation requires coordination across the organization, according to theCUBE Research’s Rob Strechay.
“The organizations succeeding today are focusing less on models and more on operationalization,” Strechay said. “They’re aligning data teams, infrastructure teams and business stakeholders around measurable outcomes and repeatable deployment strategies.”
During the Supermicro Open Storage Summit interview series, industry leaders examined the evolution of storage architectures to meet those production demands. Their discussions explored what it takes to build an efficient, adaptable foundation for enterprise AI. (* Disclosure below.)
Here are three insights you may have missed from theCUBE’s coverage of the event:
Insight #1: AI storage strategy changes the economics of inference.
Keeping graphics processing units productive requires storage that can serve active workloads quickly while accommodating growing volumes of less frequently accessed data. An effective AI storage strategy balances those demands across performance and capacity tiers, pointed out Greg DiFraia, senior vice president of AI and alliance partnerships at Scality Inc.
“We need to drive utilization up as high as it can go, and in order to do that, you’ve got to really think about things differently,” he said. “Because when we look at the topology for customers, it’s that the life cycle, when you’re talking about tens or hundreds of petabytes or even exabytes, it’s not all going to live in flash.”
Expanding agent contexts are also creating demand for storage that stores intermediate calculations used during inference. When those key-value caches outgrow GPU memory, additional tiers must balance capacity with fast access. Solidigm Inc.’s flash drives, Super Micro Computer Inc.’s integrated systems and Vast Data Inc.’s data platform support this evolving hierarchy, emphasized Anat Heilper, director of AI architecture at Vast Data.
“When we talk about [key-value] cache, which is a very significant optimization that can be done in AI inferencing … in essence, it’s the ability to replace compute with storage,” she told theCUBE. “When you have very high KV cache hit rates, we both save on compute and reduce the latency significantly.”
Insight #2: Production AI requires workload-specific systems and operational controls.
Production infrastructure must support the business decisions organizations need to make. In financial services, the speed of data processing can affect risk assessment and capital availability, noted Moiz Kohari, vice president of enterprise AI and data intelligence at DataDirect Networks Inc., who spoke alongside Vince Chen, senior director of solutions architecture at Supermicro.
“If you’re a company like BlackRock or State Street where you’re looking at a trillion dollars on your books and you have 6% to 12% capital lockup, that’s a big deal,” Kohari said. “You’re carrying that risk on your books. We can do those calculations because we can move that data back and forth quickly. When you do that, you’re unlocking 6% of that capital, so that capital can be leveraged to do anything else.”
Meeting those requirements also means assembling components that work together at the required scale. Supermicro’s partner approach addresses that deployment burden through integrated configurations, according to Chen.
“For the enterprise customers to build the AI factory as an AI purpose-built infrastructure, there has been complexity in that,” he said. “What we are doing here is working with our partners to build vertically integrated and then pre-validated T-shirt size options for our customers to choose from.”
Deployment is only the beginning of the operational challenge. As pilots become production services, organizations must accommodate broader usage within infrastructure limits, explained Ruhi Sehgal, agentic AI solutions marketing lead at Nutanix Inc.
“But with this shift comes a new dynamic we know all too well, which is we have highly constrained resources that need to be shared by a massive set of new users,” she said. “How do I ensure resilience? What infrastructure admins do every day — which is managing multi-tenancy, ensuring security [and] optimizing performance — that’s exactly what AI needs.”
Insight #3: Enterprise data becomes useful through access, orchestration and governance.
An AI storage strategy must make enterprise data usable across applications while keeping it available. Lakehouse architectures built around open formats provide a foundation for connecting traditional analytics and newer AI workflows. MinIO Inc.’s AIStor supports Apache Iceberg tables and the Iceberg REST catalog, while Advanced Micro Devices Inc.’s processors and Supermicro systems supply the underlying compute, memory and connectivity. Bringing those capabilities together gives organizations more flexibility in putting their data to work, pointed out Greg DeMichillie, vice president of product and technical marketing at MinIO.
“On the data lake side, I think it’s hard to overstate just what a big deal Iceberg has been,” he told theCUBE. “If you think back, it’s not that long ago that every database had its own proprietary format. The idea of an open table format was just bonkers … but we are really seeing now the incredible widespread adoption. That’s enabled us to do really clever things like integrating support for Iceberg into the storage tier.”
That flexibility must extend to documents, images, logs and other unstructured information spread across the enterprise. Coordinating access and movement across the data estate requires several complementary layers: Supermicro’s validated systems, Hammerspace Inc.’s unified namespace and data orchestration, Cloudian Inc.’s object storage and Seagate Technology LLC’s capacity for longer-term retention. Together, those components support a data life cycle spanning active AI use and archival storage, emphasized Sherry Lin, senior product manager of SDS solution at Supermicro. Governance remains essential throughout that life cycle, noted Peter Sjoberg, vice president of worldwide solution architects at Cloudian.
“We see a key goal to put that unstructured data under management so that it is protected, it is safe and secure,” Sjoberg said. “That’s exactly what we’re doing now as this data has moved into the AI era. That data will need to move for different purposes, and you will need to make sure it’s always under your control, unstructured or otherwise, so that you can bring it to your AI needs.”
Get more perspectives on storage and enterprise AI:
- Guests from Supermicro, Vast Data, Kioxia Holdings Corp. and Crusoe Inc. discuss their collaboration, the constraints of current enterprise infrastructure and how agentic AI is reshaping the industry.
- Guests from Supermicro, IBM Corp. and Kioxia talk about how workload requirements, data architectures and coordinated infrastructure design shape inference performance at scale.
- Guests from Supermicro, Hammerspace and Sandisk Corp. discuss how organizations are modernizing storage environments, consolidating infrastructure and building foundations capable of supporting both current and future AI workloads.
- Guests from Supermicro, Solidigm and DDN discuss how integrated AI data platforms can simplify inference deployment, support data control and expand with demand.
- Guests from Supermicro, Scality, Intel Corp. and Iron Mountain Inc. discuss how hybrid storage architectures are evolving, what service providers are learning from customer deployments and how organizations can balance agility with operational control.
- Guests from Supermicro, AMD and Nutanix Inc. talk about their collaboration and the practicalities of AI adoption.
To watch more of theCUBE’s coverage of the Supermicro Open Storage Summit interview series, stay tuned for our complete video playlist.
(* Disclosure: TheCUBE is a paid media partner for the Supermicro Open Storage Summit interview series. Neither Supermicro, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)