Optimizing Network Performance for AI Workloads: Nutanix and NVIDIA's Collaborative Approach

AI workloads require high-performance, secure, and reliable network infrastructure to operate efficiently. The increasing complexity and scale of AI applications demand optimized network stacks to prevent bottlenecks and ensure seamless data processing. The Nutanix Cloud Platform solution contains highly optimized network capabilities that are designed to meet these demands, providing a robust foundation for modern workloads and AI applications.

The Nutanix Flow solution - a component of the Nutanix Cloud Infrastructure offering - provides software defined virtual networking and network security capabilities. It enables customers to abstract underlying physical networks and introduce a consistent security posture across all workloads and locations. This allows comprehensive security and compliance support. Nutanix Flow facilitates self service consumption and secure multi-tenancy,  thus fostering seamless workload mobility without lock-in nor any re-architecting or re-IPing.

Flow comprises of:

• A logically centralized control and management plane housed within the Nutanix Prism Central multi-cluster manager, and
• A fully distributed forwarding and enforcement plane instantiated in software in the kernel of every AHV compute host.

NVIDIA offers several networking optimizations—centered around the NVIDIA BlueField line of data processing units (DPUs)—that are designed to offload and accelerate networking and data movement tasks from host CPUs, and enable secure, high-performance AI and data center operations.

At the core, NVIDIA BlueField DPUs integrate programmable Arm cores, high-speed network interfaces, and acceleration engines for functions such as encryption, storage virtualization, telemetry, and packet processing. This offloading frees the host CPU, GPU and memory resources to focus on compute-intensive workloads like AI inference or training. In AI data centers, NVIDIA BlueField DPUs also enable software-defined, zero-trust networking, isolating workloads and enforcing security policies in hardware. Through integration with NVIDIA DOCA, developers can program Bluefield for custom data-path optimizations, monitoring, or AI-aware networking. 

Additionally, NVIDIA ConnectX SuperNICs deliver low-latency, high-throughput AI compute networking—critical for GPU clusters used in AI factories and supercomputing environments.

Together, these technologies transform traditional networks into accelerated, intelligent fabrics—capable of scaling massive GPU clusters efficiently while maintaining performance and security.

Nutanix is working with NVIDIA to offload the Flow networking dataplane to NVIDIA's smart network adapters, including NVIDIA ConnectX and NVIDIA BlueField. These offloads facilitate unprecedented network dataplane performance and secure the network dataplane while also alleviating the consumption of memory and processing resources on the compute host required for network infrastructure purposes.

Integration of NVIDIA BlueField and ConnectX SuperNICs complement the scale-out networking provided by the NVIDIA Spectrum-X Ethernet switch family (validated as Nutanix Ready) which offers exceptional density, power efficiency and low latency at enterprise grade reliability.

Such performant, secure and reliable network stacks are rapidly becoming critical and indispensable components of modern AI stacks.

Future-Proofing with Latest NVIDIA BlueField Generations

Saveen Pakala, Vice-President of Product Management at Nutanix, emphasized the company's commitment to supporting the latest generations of NVIDIA BlueField. "As AI continues to drive innovation, we're dedicated to ensuring our customers can leverage the latest advances to optimize their infrastructure with a full-stack solution that provides quick ROI. Our collaboration with NVIDIA on the latest NVIDIA BlueField generations underscores our commitment to delivering optimal performance, security, and reliability for AI workloads through our platform," Pakala stated.

“Enterprise AI factories must be agile—adapting in real time to evolving workloads, ensuring data moves securely, and delivering programmable performance at scale,” said Justin Boitano, vice president of enterprise AI products at NVIDIA. “With NVIDIA BlueField integrated into software-defined platforms like Nutanix Flow, enterprises can architect secure, dynamic, and high-throughput AI infrastructures—ready to meet the demands of agentic computing and continuous inference.”

“AI requires accelerated infrastructure to build AI factories that convert data into knowledge,” said Kevin Deierling, SVP Networking & Storage at NVIDIA. “The integration of BlueField-4 with software-defined networking platforms like Nutanix Flow, provides customers with powerful new capabilities—zero-trust isolation, rapid data movement, and programmable control across the entire AI compute fabric, enabling organizations to scale securely, efficiently, and with full visibility into their infrastructure.”

Benefits to Customers

The integration of Nutanix Flow with NVIDIA BlueField DPUs offers several benefits to customers:

• Enhanced Performance: Optimized network performance for AI workloads, ensuring faster data processing and reduced latency.
• Improved Security: Secure network dataplane operations, protecting sensitive data and applications.
• Resource Efficiency: Reduced consumption of compute host resources, allowing for more efficient use of host infrastructure for application workloads..

Nutanix Cloud Platform is highly optimized to support all kinds of modern and AI workloads on-premises, in public clouds, and at the edge. To learn more about how Nutanix can help you jumpstart your AI initiatives, visit https://www.nutanix.com/enterprise-ai or contact your Nutanix account manager or channel partner.

 _{©2025 Nutanix, Inc. All rights reserved. Nutanix, the Nutanix logo and all Nutanix product and service names mentioned are registered trademarks or trademarks of Nutanix, Inc. in the United States and other countries. NVIDIA, the NVIDIA logo, and the NVIDIA product names mentioned  are registered  trademarks or trademarks of NVIDIA Corporation in the U.S. and other countries. All other brand names mentioned are for identification purposes only and may be the trademarks of their respective holder(s).}