F5 and NVIDIA Boost AI Inference Efficiency With 40% Token Throughput Gain — F5, Inc.

Event summary

F5 and NVIDIA expanded their collaboration to optimize AI inference infrastructure, combining F5 BIG-IP Next for Kubernetes with NVIDIA BlueField-3 DPUs.
The integration delivers up to 40% increase in token throughput, 61% faster time to first token (TTFT), and 34% reduction in latency.
The solution enables secure multi-tenant AI platforms at scale, addressing key metrics like token economics, GPU utilization, and cost per token.
Testing validated by The Tolly Group confirms performance gains without requiring model modifications.
The partnership aims to transform AI factories into efficient, monetizable platforms for the agentic AI era.

The big picture

The collaboration between F5 and NVIDIA addresses the critical need for infrastructure efficiency in AI systems, where success is increasingly measured by token economics and sustained throughput. As enterprises shift from AI experimentation to revenue-generating services, optimizing GPU utilization and reducing latency becomes paramount. This partnership positions F5 BIG-IP Next for Kubernetes as a strategic control plane for AI factory economics, enabling organizations to maximize the economic output per accelerator.

What we're watching

Infrastructure Efficiency: How sustained gains in token throughput will impact AI factory economics and revenue per GPU.
Market Adoption: The pace at which enterprises and GPUaaS providers adopt this solution to optimize their AI inference workloads.
Competitive Dynamics: Whether F5 and NVIDIA can maintain their lead in AI infrastructure optimization amid rising competition.