F5 and NetApp keep AI data moving from storage to GPUs

Jesse SummanProduct Marketing Manager | F5

Enterprise AI takes more than faster GPUs and bigger models. As storage teams stand up capacity for training, fine-tuning, retrieval-augmented generation (RAG), and inference, they hit a physical limit, the speed of moving massive datasets between storage and compute. That limit shows up in the metrics storage leaders answer for, missed SLAs, rising cloud storage spend, and capital requests that are hard to defend.

Great storage is only half the job

Most AI architectures rely on S3-compatible object storage to hold everything from raw training sets and vector databases to final model weights. At scale, this storage environment must handle relentless, concurrent read and write demands. While a few milliseconds of latency seems negligible on paper, it multiplies quickly across billions of objects and thousands of requests. Slow data delivery starves compute, stretches training runs, and pushes inference response times past the service levels storage teams commit to.

In short, your AI performance is bound by the efficiency of your data path. When network bottlenecks, overloaded storage nodes, or legacy security policies restrict traffic, expensive hardware sits idle. Great storage is essential, but its value only shows up when data reaches the workload quickly and securely. That gap is where SLA misses, idle GPUs, and unjustifiable storage costs come from.

The hidden bottleneck in AI data delivery

At scale, the constraint is rarely the model. It is data movement.

Training, fine-tuning, and inference workloads pull constantly from S3-compatible object storage. As datasets expand, the storage layer must keep pace, serving files without dropping throughput or availability. If that data stream slows, the entire downstream pipeline stalls.

Four bottlenecks come up again and again:

• Storage node congestion. Uneven traffic across object storage clusters overloads individual nodes, while others sit idle.
• Transport-layer limits.Network throughput ceilings and TLS processing on storage controllers throttle data moving between storage and compute pools.
• Security overhead.Cumbersome inline checks slow pipelines without improving real-world protection.
• Storage cost gravity. Datasets pinned to expensive public cloud tiers inflate operating costs and make on-premises capacity harder to justify.

The impact reaches well past storage metrics. Slowdowns drag out training runs, spike inference response times, and leave expensive GPUs idle. Friction in the data path erodes the return on storage and compute you have already paid for.

Keeping pipelines at full capacity takes more than a large bucket. It takes a data delivery path built to feed compute clusters quickly, securely, and without interruption.

How F5 and NetApp create a resilient data path

Scaling AI workloads takes high-performance storage and an efficient path to route that data to compute.

NetApp StorageGRID handles the massive volumes of S3-compatible object data required for training and inference, while F5 BIG-IP sits in front of StorageGRID to manage, secure, and accelerate that traffic. BIG-IP acts as a high-throughput delivery controller, optimizing how data moves from storage pools to GPU clusters.

Together, F5 and NetApp resolve the data-path bottlenecks storage teams own. Among the benefits include:

• Performance and throughput. BIG-IP balances storage requests, steers traffic, and prioritizes critical workloads, preventing hot spots on individual storage nodes. Hardware-accelerated TLS termination can keep cryptographic load off storage controllers, so bulk S3 transfers run closer to wire speed.
• Availability and visibility. The joint solution keeps data flowing during storage node failures, routine maintenance, or unexpected traffic spikes. At the same time, F5 telemetry and Layer 7 visibility help infrastructure teams monitor traffic patterns, identify usage trends, and resolve performance bottlenecks before they impact AI workloads.
• Pipeline security. BIG-IP protects the data path with hardware-accelerated encryption and access controls, protecting sensitive training datasets and proprietary model weights without degrading transfer speeds.
• Storage flexibility and cost control. BIG-IP acts as a storage-agnostic front door, intelligently directing S3 requests to the optimal StorageGRID cluster based on performance, availability, or policy. Whether organizations operate a single deployment or multiple StorageGRID environments globally, F5 helps ensure users and AI workloads connect to the best storage resource without application changes.

The result is a fast, reliable, and secure path between object storage and compute. Removing transit delays keeps GPUs busy, holds storage service levels under bursty AI load, and gives storage leaders a clearer return on the infrastructure they already own.

The F5 and NetApp deployment guide is now available

Knowing the value of resilient AI data delivery is one thing. Standing it up in production is another.

That is why F5 and NetApp have published a joint deployment guide, available now. It details how to configure BIG-IP in front of StorageGRID to optimize S3 traffic, maintain high availability across storage nodes and sites, and secure the data path end to end. Storage and infrastructure teams will find validated architecture designs, step-by-step configuration for virtual servers, S3 health monitoring, and TLS offload, plus tuning guidance for high-volume object transfers.

The joint deployment guide serves as a practical blueprint for engineering teams building data pipelines for training, inference, or RAG. Instead of designing an S3 front door from scratch, teams can deploy a tested pattern, prove the throughput and availability numbers in their own environment, and use that evidence to defend the next capacity decision.

Download the F5 and NetApp deployment guide to learn more.

Meet the F5 team at NetApp INSIGHT 2026

F5 is a proud sponsor of NetApp INSIGHT 2026, which starts today and runs through October 1 at the MGM Grand in Las Vegas, Nevada. Our team will be on the floor with the joint F5 and NetApp architecture, ready to talk through the data-path problems storage teams are living with, including node congestion, TLS offload, multi-site failover, and moving AI datasets back on-premises.

Visit F5 NetApp INSIGHT at booth #303 and add our session, “Why data delivery matters and how F5 and NetApp enable resilient storage,” on Wednesday, Sept 29, at 2:30 pm, to your INSIGHT agenda. We’ll see you in Las Vegas.

About the Author

Jesse SummanProduct Marketing Manager | F5

More blogs by Jesse Summan