NVIDIA AI Podcast: How AI Data Platforms Are Shaping the Future of Enterprise Storage

Episode 281 | November 18, 2025
Host: Noah Kravitz (A)
Guest: Jacob Lieberman (B), Director of Product Management, NVIDIA Enterprise Product Group

Episode Overview

This episode explores how AI data platforms—specifically, GPU-accelerated storage—are transforming enterprise data management, security, and AI-readiness. Host Noah Kravitz and NVIDIA’s Jacob Lieberman delve into the practical and technical challenges of deploying AI agents at enterprise scale, the importance of in-place, secure data transformation, and the industry-wide shift toward bringing compute directly to the data. The conversation draws parallels with historical transitions in storage and paints a vivid picture of future enterprise infrastructure.

Key Discussion Points & Insights

1. Agentic AI Adoption in the Enterprise

• Consumer vs. Enterprise Needs:
  Many people now use AI agents in daily life. However, “the needs of a consumer are very different than the needs of an enterprise.” (B, 01:47)
• Rapid Technological Progress:
  “The open models of today are just as powerful as the commercial models were a little while ago, but it’s still challenging for enterprises to put agents into production.” (B, 02:04)

2. Major Challenges to Enterprise AI Adoption

• Bridging Proof of Concept to Production:
  Enterprises are eager to increase AI adoption, but “it’s difficult to move from a POC state to production...there are still challenges.” (B, 02:36)
• Data Access Issues:
  All AI systems require “secure access to accurate recent data,” which is complicated by the heterogeneity and unstructured nature of enterprise data. (B, 03:02)
• Unstructured Data:
  Most enterprise data—like presentations, PDFs, media files—lacks structure, making it difficult to feed into AI pipelines.

  “The vast majority of enterprise data is unstructured… you can’t just shove into a database and query with a structured query language.” (B, 04:14)

3. AI-Ready Data and Data Transformation Pipelines

• Making Data AI-Ready:
  Establishing pipelines that:
  • Find and gather data
  • Extract semantic knowledge
  • Chunk data into manageable pieces
  • Enrich with metadata
  • Embed into numerical (vector) representations
  • Index into vector databases (B, 05:11)
• Data Velocity and Governance:
  Continuous process required due to the combined effect of new and changing data—termed “data velocity.” Not knowing what’s changed makes enterprises re-process vast datasets unnecessarily.

  “Enterprises are…rewashing all their dishes every time they have one dirty dish.” (B, 07:48)

4. Data Security and Data Drift

• Security Pitfalls in Copy-Based Pipelines:
  Making data AI-ready often means making multiple copies, which increases risks:
  • Greater attack surface
  • “Disconnected from the source of truth”—not tracking changes or permission updates

  “What if Noah should no longer have access to this document? …but he can still access all the copies.” (B, 09:11)

5. Enter the AI Data Platform

• NVIDIA’s Reference Design:
  The AI Data Platform (AIDP) offers a blueprint for partners to transform storage with GPUs, making data AI-ready in place and continuously, avoiding unnecessary copies and movement. (B, 09:36)
• Key Innovation:
  “Instead of sending all your data to the GPU, you can actually send your GPU to the data.” (B, 11:48)
• Data Gravity:
  Enterprises prefer to keep data in place due to size, cost, and compliance—so moving compute to storage is logical.
• Operational Impact:
  “Letting [GPUs] operate on the data in place where it lives” enables instant updating of representations and permissions whenever the source changes. (B, 11:48–12:30)

6. Challenges & Industry Mindset

• Cultural and Technical Pushback:
  Some in traditional storage roles find the idea radical.

  “AI will be everywhere… in your storage? No way. That will never happen. Storage hasn’t changed in 30 years.” (B, 13:22)

7. How GPUs Are Used in AI Data Platforms

• End-to-End Intelligent Data Operations:
  The GPU is utilized to:
  • Extract and process data continuously
  • Embed and index content
  • Perform semantic search “not keywords in the documents, the GPU can be used for all of those things…” (B, 14:31)
• Background Operations:
  These processes happen automatically, freeing data scientists to focus on higher-value tasks. “Maybe some estimate up to 80% of their time [data scientists spend] wrangling data...this frees up your precious data science resources to actually do data science.” (B, 15:18)

8. NVIDIA’s Reference Design Model and Industry Impact

• Partner-Centric Approach:
  NVIDIA provides reference designs and AI blueprints; storage partners innovate and differentiate on top.

  “The storage industry is eager to transform…they are taking these designs and…innovating in many ways we didn’t initially expect.” (B, 17:50)

• Agent Use in Storage:
  Novel use-cases emerge such as:
  • Agents classifying sensitive documents in storage
  • Agents monitoring system telemetry for optimization (B, 19:01–19:45)

9. “Letting Your AI Agents Work From Home” Metaphor

• Efficiency through Localized Compute:
  “It’s more efficient to send the compute to the data than to send the data to the compute.” (B, 21:04)
  • Just as remote work eliminates the commute, so does localized compute for data processing.

10. NVIDIA’s Role in the Storage Ecosystem

• Better Together:
  Storage partners handle protection/governance, NVIDIA accelerates.

  “Now, by partnering, we can both bring to the table what we’re uniquely good at…better than what we could do by ourselves.” (B, 23:28)

11. Metaphor: The Librarian and Enterprise AI

• Analogy:
  Librarians (agents) who know the library (data) can find and suggest more relevant information, as long as their knowledge is current and comprehensive. (B, 27:08)
• Reality:
  AI Data Platforms are live with lighthouse customers, with continuous feedback and rapid evolution in the field. “I imagine a future where all storage will be GPU accelerated and there will be no AI factory that does not have an AI data platform.” (B, 27:59)

12. Evolution, Not Revolution

• From CPUs to CPUs+GPUs in Storage:
  Today’s move is an evolution akin to the shift from local server storage to network-attached storage using CPUs. Now, GPUs take on the next level:

  “Instead of just serving the data and protecting the data, they [storage CPUs] were doing things with the data like compression, decompression, encryption, and keyword search…So if we extend that now with a GPU…we have embedding and indexing and semantic search.” (B, 29:22)

13. AI Data Platform in the AI Factory

• Separation of Concerns:
  Decouple compute and data scaling, enabling cost-effective, flexible resource management.
• Faster Innovation:
  Compute (AI models and software) can evolve rapidly while the data and its protections remain stable and secure. (B, 32:49)

Notable Quotes & Memorable Moments

• On Data Gravity and Compute:
  “Instead of sending all your data to the GPU, you can actually send your GPU to the data.” (B, 11:48)
• On Data Security:
  “Every time you make a copy, you’re increasing the attack surface on your data.” (B, 08:28)
• On the Move to AI in Storage:
  “AI will be everywhere… in your storage? No way. That will never happen. Storage hasn’t changed in 30 years.” (B, 13:22)
• On Efficiency:
  “It’s more efficient to send the compute to the data than to send the data to the compute.” (B, 21:04)
• Vision for the Future:
  “I imagine a future where all storage will be GPU accelerated and there will be no AI factory that does not have an AI data platform. There’s too many benefits to it.” (B, 27:59)
• On Evolution in Storage:
  “The divisions we’ve put up around data management for AI and regular enterprise data management are somewhat artificial. When the GPU is in the storage, that division goes away.” (B, 29:44)

Timestamps for Key Segments

• [01:46] Agentic AI adoption differences in consumer vs. enterprise
• [03:02] Challenges in enterprise AI deployment—data, legacy systems
• [04:14] Unstructured enterprise data and AI-readiness
• [05:11] Steps to make data AI-ready
• [07:08] Data velocity and continuous transformation
• [09:11] Security challenges with duplicate data
• [09:36] Introduction to the AI Data Platform and reference design
• [11:11] Moving compute to data—reducing data movement
• [13:22] Industry perceptions and resistance to storage innovation
• [14:31] How GPUs operate inside the storage stack
• [15:18] Impact on data scientists and background data preparation
• [18:34] Running agents inside storage—early examples
• [21:04] “Work from home” metaphor for agents and compute efficiency
• [23:03] NVIDIA’s collaborative approach with storage partners
• [27:08] Library and librarian metaphor; readiness of AI data platforms
• [29:22] Natural evolution of storage—CPU to GPU
• [30:43] What is an "AI factory" and the role of the AI Data Platform
• [32:49] Decoupling scaling and lifecycle management between compute and data

Where to Learn More

• Further Reading & Case Studies:
  “If you search NVIDIA AI Data Platform, that will take you to an estate on our webpage that actually has links to all of our partners’ implementations.” (B, 33:13)

Final Thoughts

Jacob Lieberman paints a compelling and pragmatic vision of enterprise AI’s future: a world where the boundary between data storage and AI compute dissolves, enabling new efficiencies, security, and intelligence. The power of GPU-accelerated storage lies not merely in optimization but in fundamentally reimagining how enterprises manage, secure, and extract value from rapid streams of unstructured data.