Powering the New Economics of AI: How DDN Fuels Nvidia’s Record-Breaking Inference Results

Blog written by DDN published Dec. 17, 2025

When Nvidia announced its latest InferenceMAX benchmark results, one fact stood out above all others: AI performance and efficiency have reached an entirely new frontier. The Nvidia Blackwell platform and GB200 NVL72 system delivered unprecedented throughput, responsiveness, and ROI, redefining what’s possible in AI inference at scale.  

At DDN, we’re proud to have played a foundational role in powering these achievements. 

For more than two decades, DDN has been the silent dominant force behind the world’s most advanced AI and HPC environments, from Nvidia’s internal AI factories  to sovereign research centers and cloud hyperscalers. The InferenceMAX results underscore not just Nvidia’s hardware and software innovation, but the data intelligence infrastructure that enables it in various AI workloads. 

The AI Factory Runs on Data Intelligence
A $5 million investment in a Nvidia GB200 NVL72 system generates $75 million in token revenue, which is a 15x ROI and is the new benchmark for AI economics. But it’s only achievable when the data layer operates as intelligently and efficiently as the accelerated compute layer. 

That’s where DDN comes in. DDN provides up to 1.5X higher GPU utilization, which means your ROI will be lower if you don’t use DDN. 

The DDN AI400X3, powered by the EXAScaler parallel file system ensures GPUs like Nvidia’s Blackwell and H100 stay fully saturated with data.  

This maximizes every watt, every token, and every cycle.  

Key MLPerf Storage v2.0 Results  

• 30.6GB/s read and 15.3 GB/s write throughput, enabling lightning-fast checkpointing for models like Llama3-8B  (3.4 seconds to load and 7.7 seconds to save) 
• 208 H100 GPUs on ResNet50 powered by a single 2U, 2400-watt appliance
• 120+GB/s sustained read throughput for Unet3D training 
• Scalability up to 640 simulated H100 GPUs on ResNet50 - over  2x improvement year-over-year

Proof point: Data performance is compute performance. When GPUs never wait for data, inference accelerates – and AI turns into a business engine. 

Efficiency is the New Performance
The Nvidia’s InferenceMAX benchmarks and DDN’s MLPerf Storage results reveal a shared truth:  

Efficiency is the new measure of AI performance.

Nvidia’s Blackwell architecture delivers 10x higher throughput per megawatt compared to the previous generation. 

In parallel, DDN’s AI400X3 achieves unmatched performance density per watt and per rack unit.  

Together, they redefine the AI data center economics – delivering faster performance, higher utilization, and lower total cost of ownership. 

And with its compact 2U form factor and low power profile, the AI400X3 enables sustainable AI growth without compromising scale.  

In an era where power and cooling are premium resources, DDN enables organizations to expand AI responsibly while maximizing ROI. 

From Benchmarks to Breakthroughs
Benchmarks matter – but what matters most is the impact in real-world deployments. 

• National Labs: At Lawrence Livermore National Laboratory, DDN’s AI storage powers one of the world’s fastest AI-enabled supercomputers – accelerating breakthroughs in materials science and clean energy research
• Autonomous Systems: Automotive leaders rely on DDN to accelerate real-time model inference and simulation pipelines at global scale
• Enterprise AI Factories: In enterprise environments, DDN integrates with Nvidia DGX Cloud and GB200 systems to orchestrate intelligent data flows across multi-tenant AI workloads, improving efficiency and economics for production-scale deployments

From HPC to AI inference, DDN ensures that data moves as intelligently as the models it fuels – turning potential into productivity. 

A Proven Partnership in AI Performance
Since 2016, Nvidia has relied exclusively on DDN to power its internal AI clusters – a testament to the reliability and scalability of our technology. That collaboration continues to evolve with each generation of innovation, from H100 to Blackwell and beyond. Whether it’s record-breaking  InferenceMAX results or MLPerf Storage leadership, the common thread is clear: every leap in compute capability is made possible by intelligent data performance.  

The Road Ahead – Turning Performance into Profits
As enterprises and nations build the digital economies of tomorrow, DDN is proud to power that transformation alongside Nvidia and our ecosystem partners, enabling organizations to turn performance into profits and data into intelligence. Because in the new economics of AI, data is multiplier, and DDN is the intelligence that makes it matter.