Networking software company F5 and Nvidia deepened their AI infrastructure partnership.

F5 Networks and Nvidia have expanded their collaboration to optimize AI inference workloads through an integration of F5's BIG-IP Next for Kubernetes with Nvidia's BlueField-3 data processing units. The partnership is designed to improve GPU utilization and accelerate token throughput for large-scale AI deployments by offloading critical networking, security, and traffic management functions from CPUs and GPUs to the specialized DPU silicon. Nvidia shares traded relatively flat on the news, down 0.16% in today's session, while declining 1.99% year-to-date. F5 Networks similarly experienced modest trading activity following the announcement, with markets treating the expanded partnership as an evolutionary step rather than a near-term revenue driver.

The expanded integration centers on running F5's BIG-IP Next for Kubernetes directly on Nvidia's BlueField-3 DPUs to create an intelligent, telemetry-aware infrastructure layer. By moving load balancing, Layer 4-7 security functions, firewall operations, DDoS protection, API security, and TLS encryption to the BlueField-3 DPU, the combined solution frees GPU resources to focus exclusively on AI computation. According to solution documentation from F5, this architectural approach aims to increase effective tokens per second per dollar of GPU investment while simultaneously reducing latency in large language model and retrieval-augmented generation workloads. The partnership specifically targets secure multi-tenant AI platforms and GPU-as-a-Service business models, where network isolation, zero-trust architecture, and tenant separation become critical operational requirements.

For Nvidia, the F5 partnership represents another step in positioning itself as a full-stack AI infrastructure provider rather than solely a GPU vendor. While BlueField DPUs currently represent a small revenue contributor compared to Nvidia's massive GPU business, they are increasingly central to the company's vision of "AI factories" where every component of the data center rack—from networking to security to storage—operates as an integrated, optimized system. The tie-up with F5, a tier-one player in enterprise application delivery and security with established relationships across Fortune 500 companies and telecommunications providers, opens distribution channels for Nvidia's DPU technology into customer segments that may have been slower to adopt AI-specific infrastructure.

From F5's perspective, the partnership addresses competitive pressure as the company transitions its customer base from legacy BIG-IP appliances to cloud-native, containerized architectures. Positioning BIG-IP Next as the security and traffic management layer for AI inference workloads creates a new growth vector as enterprises modernize data centers for generative AI applications. The announcement arrives as competition in the DPU and SmartNIC market intensifies, with AMD's acquisition of Pensando, Intel's infrastructure processing units, and custom silicon development by hyperscale cloud providers all representing alternative approaches to offloading networking and security functions in AI-optimized data centers.

Recent Nvidia announcements have highlighted expanded roles for DPUs in future architectures, including the upcoming Rubin platform and AI-native storage systems designed for long-context memory in agentic AI applications, suggesting a multi-year partnership trajectory that could extend through subsequent DPU generations. Neither company disclosed specific financial terms or revenue projections tied to the expanded collaboration. Analysts note that the direct near-term revenue impact for Nvidia from this particular integration is likely modest relative to the company's GPU-driven business, which has powered extraordinary growth over the past two years as generative AI adoption accelerated. However, the strategic value lies in building a comprehensive ecosystem where partners like F5 standardize on BlueField, creating switching costs and platform lock-in effects that extend beyond raw silicon performance.