Powering AI Factories from the Data Layer Up

AI infrastructure is undergoing a fundamental shift. What started as massive, centralized clusters built purely for training models is evolving into something far more dynamic—systems designed to deliver inference continuously, at scale. The NVIDIA AI Factory MGX Ecosystem celebrates this evolution to data centers purpose-built to produce and serve AI outcomes across a growing range of workloads.

But as these AI factories take shape, one thing is becoming increasingly clear: performance is no longer just about compute. It’s about data, how fast it moves, how reliably it’s accessed, and how efficiently it scales across the system.

Why Data Is Now the Bottleneck and the Opportunity

AI workloads today—from training to fine-tuning, and especially real-time inference—depend on constant access to massive datasets. Models must retrieve weights, context, and intermediate data like KV cache with extremely low latency and consistent throughput. When the data layer can’t keep up, even the most advanced GPUs sit idle, waiting. This is why architectures like NVIDIA MGX are gaining traction. MGX introduces a modular approach, allowing system builders to combine compute, networking, and storage in flexible ways to match specific AI workloads. In this model, storage isn’t just supporting infrastructure, it’s a core enabler of performance at scale. And Sandisk is helping define that layer.

The Role of High-Performance Storage

Modern NVMe™ SSDs sit at the center of this transformation, delivering the low latency and high throughput needed to keep AI pipelines moving. Sandisk’s enterprise portfolio, led by drives like the Sandisk® DC SN861 PCIe® Gen5 NVMe™ SSD, is designed for these environments. They provide the performance and consistency required for training and inference pipelines.

Beyond Performance: Building for Real-World AI Systems

As AI deployments scale, storage must do more than just go fast, it must integrate cleanly into increasingly dense, modular, and thermally constrained systems. This is where Sandisk continues to innovate across connectivity, thermals, and form factor evolution.

Keeping Data Moving with High-Speed Connectivity

In AI systems, data is constantly in motion—flowing between storage, GPUs, and networking layers. High-speed PCIe Gen5 connectivity helps ensure Sandisk SSDs can keep pace, feeding accelerators fast enough to minimize bottlenecks. Standard NVMe interfaces enable seamless integration across NVIDIA MGX platforms.

Managing Heat in Dense AI Deployments

AI servers are pushing the limits of density. GPUs, networking, and storage operate within tightly packed systems, increasing thermal pressure. Sandisk enterprise SSDs are engineered to maintain consistent performance under sustained workloads, even in these constrained environments.

Form Factors That Match Modular Design

E1.S has become a foundation for high-density NVMe deployments, enabling efficient airflow and serviceability at scale. Sandisk has been amongst those at the forefront of E1.S adoption in modular AI systems. 

Looking ahead, the industry is evolving toward E3.S form factors to support higher power, improved thermals, including liquid-cooled environments, and increased capacity scaling aligned with next-generation AI workloads.

Scaling Across the Ecosystem

Sandisk aligns its enterprise SSD portfolio with modular architectures like MGX, enabling deployment across a broad ecosystem of OEMs and cloud providers. This helps ensure drives like the Sandisk DC SN861 SSD can scale across diverse AI system designs in a range of capacity points.

The Bigger Picture

AI factories represent the next phase of data center evolution, where intelligence is produced continuously and at scale. By advancing enterprise SSD technology, from platforms like the Sandisk DC SN861 SSD evolving to form factors like E3.S, Sandisk is helping build the data storage foundation that AI factories depend on. 

* 1TB = 1 trillion bytes. Actual user capacity may be less depending on operating environment.