The data center infrastructure that many enterprises are deploying to support artificial intelligence (AI) resembles bringing a pair of high heels to a 100-meter sprint, according to Arrcus CEO Shekar Ayyar. "We are really under[estimating] what is needed to get to AI," he told SDxCentral.

In the context of AI, the industry talks a lot about GPU capacity and the cost of running workloads on GPUs in the cloud. "If you look at the underlying processor landscape, a lot of the focus on AI tends to be on GPUs and Nvidia," but that's limited thinking, he said. The answers to today's AI infrastructure challenges lie in networking, instead.

The CEO highlighted Arrcus' smart distributed network technology and Arrcus Connected Edge for AI (ACE-AI) as key to the vendor's unique position in the infrastructure technology landscape. "We are the network operating system that can straddle switching and routing; that can straddle cloud, telecom service providers and enterprise; that can bring efficiencies inside the core of a data center, between data center capacities and between GPU clusters; and then extend that into the edge of the network," Ayyar said.

Arrcus' technology, including its architecture, switches and routers, targets improvements in creating and deploying applications and network functions. The vendor's ACE-AI architecture, for example, promises 30-40% efficiency improvements in the network compared to other ways of doing the same thing, like through traditional networking gear vendors, he said.

Arrcus accomplishes all that through "very deep IP in our operating system" for network flow control, ingress congestion management and cloud egress cost management. Most organizations are operating in a distributed or multicloud environment, and to deploy AI applications across the entire network is no easy task.

"If you're not on a single cloud – if you're not on a single highway – every time you exit the Amazon highway or the Google highway or the Azure highway, you've got to pay a tax," he said. And in many cases, those taxes trump the cost of using GPU capacity on a per-unit basis.

To that point, the vendor differentiates itself by helping customers "put the right kind of network in place with features and functionality built into that network operating system," Ayyar said. Managing ingress and egress "is going to be very important, and that is the kind of technology that we are building."

The vendor claims to set itself itself apart with its networking-based approach to AI. "Every large company in the AI space has a hammer, and to them, every AI problem looks like a nail," Ayyar said. Each hyperscaler claims it has the best cloud for AI, and for network processor companies like Broadcom, "the idea is how can you take their chipsets and then build the best networks possible out of that." For Nvidia, it's all about the GPUs and processing. "We stitch this world together," he claimed.

"We can take our multicloud networking fabric and use that to connect between Azure Cloud, Google Cloud, or – for that matter – Oracle Cloud, so that the enterprise customer using [any] combination of clouds can have common visibility across their network," he said.

The CEO also described the company's operating system as compatible with multiple different processor architectures, including Broadcom, Nvidia and Intel. "The net of this is that Arrcus is differentiated as one common network operating platform that can tie together these diverse vendor processor environments," Ayyar said.