With Vast Data, the collaboration will offer optimized AI infrastructure combining high-performance networking and scalable data storage and access. The partnership pairs VAST’s Disaggregated Shared-Everything (DASE) data platform architecture with Arista’s low-latency, high-bandwidth ethernet switches.

The Nvidia partnership involves the use of Nvidia hardware including the BlueField-3 SuperNICs. Arista is now also building out a new AI Agent technology to help network administrators manage and configure complex network topologies.

Arista building out AI Agent technology

Arista’s AI Agent software is based on the EOS network operating system. This agent facilitates direct communication between the network infrastructure and server/accelerator hosts like NVIDIA’s BlueField-3 data processing unit (DPU).

“The Arista EOS-based AI Agent is a small piece of software that is installed either directly onto a SmartNIC or in servers to control the NIC,” Brendan Gibbs, Technical Director, Customer Engineering at Arista told SDxCentral. “This is not an AI Chatbot nor developed with any Nvidia software, but is rather Arista software to manage configuration of NICs in tight coordination with the network, to create optimum performance for AI network infrastructure – now extended to include NICs inside servers.”

Gibbs noted that Arista’s collaboration with Nvidia revolves around the Arista EOS-based AI Agent being able to be installed onto Nvidia Bluefield-3 SuperNICs. That integration provides remote control and visibility into that NIC to ensure it is configured correctly in concert with Quality-of-service (QOS) settings in the network for proper end-to-end AI optimization.

He explained that the AI agent streams telemetry data from the NIC upstream into a central Arista NetDL (network data lake) repository, from which Arista Cloudvision network automation software can provide analytics and reporting for network management.

The AI Agent doesn’t replace anything

The Arista EOS-based AI Agent provides an extra level of control and visibility to network operators. Gibbs emphasized that it doesn’t replace any existing Arista service.

He noted that today, network operations teams and server operations teams that are tasked with managing NICs, servers, and GPUs, tend to be separate, which can lead to potential misconfiguration between NICs and the network and/or lack of full insight to problem areas across the full AI ecosystem.

“The extra visibility and control provided by the EOS-based AI Agent provides a new level of insight and control to the network operations team who might previously have had oversight of the network alone,” Gibbs said.

VAST Data brings AI Data pipelines to Arista

On the data side, Arista is partnering with VAST Data to integrate their scalable, high-performance Disaggregated Shared-Everything (DASE) architecture into the optimized AI fabric.

The combination of VAST’s global data platform with Arista’s high-speed, low-latency networking provides enterprises with resilient data access and accelerated GPU computing capabilities. It enables consolidating enterprise workloads alongside data-hungry AI onto a unified, multi-tenant infrastructure.

“Today’s new AI workloads require a new type of data infrastructure, built on top of a radically different architecture than before,” John Mao, VP of Strategic Alliances at VAST Data, wrote in a statement.