How a 26-year-old storage company became the invisible infrastructure powering 700,000+ GPUs and earning validation from Jensen Huang himself
When Jensen Huang, NVIDIA’s CEO, sits down for what’s supposed to be a 20-minute fireside chat and instead talks for 80 minutes, you know something important is happening. That’s exactly what occurred at DDN’s Beyond Artificial event earlier this year, where Huang delivered what might be the ultimate technical validation in the AI space: “NVIDIA is powered by DDN…without DDN, NVIDIA supercomputers wouldn’t be possible.”
For most people, DDN operates in the shadows-the invisible infrastructure layer that makes AI magic happen. But during our recent briefing at the 62nd IT Press Tour in Palo Alto, CA, it became clear that this 26-year-old company has positioned itself at the epicenter of the AI revolution. With 700,000+ GPUs under management and an $800 million booking year, DDN isn’t just riding the AI wave-they’re building the foundation that makes it possible.
From Supercomputers to AI Factories: The Scale Revolution
The numbers Paul Bloch, DDN’s President and Co-founder, shared during our briefing are staggering. “Today, 700,000 [GPUs]. If you look at it, in two years, you’re probably gonna quadruple that,” he explained. But here’s what makes this even more remarkable: the fundamental shift in buying patterns.
“If you look at even two, three years ago, when people would buy eight GPUs or 32 GPUs or 200 GPUs, it was a lot. And now people buy 20,000 GPUs and nobody cares,” Bloch noted with a mix of amazement and matter-of-fact acceptance.
The largest cluster DDN has deployed? A mind-bending 200,000 GPUs at Elon Musk’s xAI facility in Memphis. “If you visit Memphis, it’s like you see in the future. It’s like literally rows and rows and rows and rows and rows and rows,” Bloch described. This isn’t just a data center-it’s what he calls one of the first true “AI factories.”
These investments dwarf traditional supercomputing. Where a $500 million machine once represented the pinnacle of high-performance computing, today’s AI deployments routinely reach 1-3 billion. A billion dollars, Bloch explains, gets you roughly 12,000-18,000 of the latest GPUs-and that’s just “table stakes” for foundation model development.
Two-Platform Strategy: Performance Meets Intelligence
DDN’s answer to this AI infrastructure challenge comes through two complementary platforms that address different aspects of the data intelligence puzzle.
EXAScaler: The Performance Workhorse
EXAScaler represents DDN’s evolution of their HPC heritage into AI-optimized infrastructure. This parallel file system has been proven at genuine exascale, handling tens of thousands of GPUs while delivering sustained performance that it says competitors simply can’t match at scale.
The platform’s latest innovations include support for NVIDIA’s Spectrum-X networking, delivering up to 3.2x read acceleration and 2.4x write acceleration. With new Blackwell GPU reference architectures, EXAScaler achieves over 90% of maximum network performance for both reads and writes-numbers that translate directly to improved GPU efficiency.
But here’s where DDN’s HPC heritage pays dividends: reliability and manageability. “AI customers don’t understand. If my system is down. Get it back on,” Bloch explained, contrasting this with traditional HPC users who would accept downtime for maintenance. EXAScaler now supports zero-downtime upgrades and enterprise-grade multi-tenancy-features that have become table stakes for AI deployments.
Infinia: The Cloud-Native Intelligence Platform
While EXAScaler handles the heavy lifting, Infinia represents DDN’s boldest bet on the future of AI infrastructure. Built from scratch as a software-defined platform, Infinia is designed for the cloud-native, multi-modal AI world that’s rapidly emerging.
The performance numbers are eye-opening: 100x faster object listing compared to public cloud providers, 25x faster time-to-first-byte response, and the ability to deliver sub-millisecond latencies at massive scale. CTO Sven Oehme explained the architectural advantage: “Every device is a key value store, and then it’s abstracted by a software development kit that directly natively implements S3. So basically, the S3 protocol just passes through and writes it into the key value store. There is no indirection layers.”
This clean-sheet design eliminates the software layer inefficiencies that plague traditional storage systems trying to bolt AI capabilities onto legacy architectures.
Real-World Performance: Where Theory Meets Reality
The proof, as they say, is in the performance pudding. DDN’s demonstrated results at GPU Technology Conference included a 22x speedup for RAG (Retrieval-Augmented Generation) pipelines-achieved not through specialized hardware, but by replacing AWS S3 with Infinia in an otherwise identical configuration.
Jump Trading provides a compelling case study in real-world performance. As their CTO Alex Davies noted, “DDN QLC systems are a really important part of that environment to get IO to our researchers as quickly as possible.” The results? A 10x decrease in processing latency that enables the parallel processing workflows essential to high-frequency trading algorithms.
In genomics, Roche Sequencing Solutions saw their analysis turnaround times drop from 15 days to just 2 days-a 7x improvement that Director of Accelerated Computing Chuck Seberino described succinctly: “The solution DDN had, for what we needed, just made sense. We needed a small footprint and something very fast so we could process the data as quickly as possible-time is money.”
The Infrastructure Efficiency Edge
One of DDN’s most compelling advantages becomes apparent when you look at deployment complexity. As Oehme illustrated during our briefing, a competitor system supporting 4,000 GPUs requires extensive networking infrastructure, thousands of cables, and multiple racks of storage equipment. DDN’s approach? Twenty appliances connected directly to the GPU infrastructure.
In this example, with DDN, you basically have, let’s say, 40 cables. And with the competitive solution, you have 1,400 cables, Bloch explained, highlighting the simplicity difference. This isn’t just about aesthetics-it translates to reduced power consumption, smaller data center footprints, and dramatically simpler management.
The efficiency gains compound at scale. Where competitors might need 10x the hardware to achieve equivalent performance, DDN’s claims their architecture delivers the same results with a fraction of the infrastructure overhead.
Market Evolution: From Training to Inference
DDN’s leadership recognizes that today’s AI landscape represents just the beginning. “Right now, it’s all about training foundational models, but actually, where the thing is really going to start taking off is when people are going to move completely into inference,” Oehme explained.
This transition from training to inference deployment brings new challenges and opportunities. Inference workloads demand consistently low latency rather than peak throughput, and they need to scale across diverse deployment environments-from on-premises data centers to public clouds to edge locations.
DDN’s cloud strategy reflects this evolution. Their first-party integration with Google Cloud, where Google offers DDN EXAScaler as its parallel file system solution, demonstrates how traditional infrastructure vendors are adapting to hybrid and multi-cloud realities. Google is now offering DDN EXAScaler to their customers for AI workloads, Bloch noted, emphasizing that this isn’t a marketplace listing but a fully integrated Google service.
The Blackstone Validation and Financial Foundation
Perhaps the strongest validation of DDN’s market position came in January 2025, when Blackstone invested $300 million for a $5 billion valuation. This wasn’t a typical venture capital round-Blackstone conducted extensive due diligence, visiting DDN’s facilities and interviewing their top 50 customers.
As Bloch described it: They didn’t leave any stone unturned. They visited all of the DDN sites. Spoke to their top 50 customers. They had multiple people doing due diligence. It was like an acquisition.
The financial foundation this provides goes beyond just capital. DDN has maintained profitability “literally forever,” as Bloch puts it, growing conservative while building sustainable technology advantages. With north of 20% EBITDA margins and expectations for over $1 billion in bookings this year, DDN operates from a position of financial strength that’s rare in the AI infrastructure space.
Looking Ahead: The Data Intelligence Revolution
As our briefing concluded, it became clear that DDN sees itself at an inflection point. The company that spent its first 20 years perfecting high-performance storage for the most demanding technical computing applications now finds itself essential to what Bloch calls a “once-in-a-lifetime situation” in AI.
The roadmap ahead includes expanded SQL capabilities for metadata-rich analytics, enhanced integration with frameworks like Apache Spark, and advanced multi-tenancy features that can dynamically adjust performance, capacity, and resilience on a per-tenant basis. These aren’t just incremental improvements-they represent DDN’s evolution from infrastructure provider to data intelligence platform.
With 8 of the top 10 cloud providers relying on DDN technology and Jensen Huang’s public endorsement, DDN has positioned itself as the invisible foundation enabling the AI revolution. In a world where data storage represents only 2-8% of AI infrastructure costs but can make or break the entire system’s performance, DDN believes their role becomes indispensable.
As AI workloads continue their relentless scale trajectory-toward the million-GPU clusters that industry leaders envision-DDN’s combination of proven technology, financial stability, and deep AI expertise positions them not just to participate in the data intelligence revolution, but to power it.
##
