Companies are pouring billions of dollars into the data centers needed to support artificial intelligence, and Ayar Labs Inc. wants to help them pay off that investment.

Ayar Labs specializes in optical interconnectivity technology for AI infrastructure. Its primary product, Optical I/O, enables low latency, high bandwidth and better energy efficiency for GPU fabrics, according to Vladimir Stojanovic (pictured), co-founder and chief technology officer of Ayar Labs.

Ayar Labs’ Vladimir Stojanovic discusses how the company is scaling up AI systems.

“The last 10 years of AI were about models and compute engines,” Stojanovic said. “The next 10 years will be about how do you interconnect all these most efficiently to be able to follow the trend of increased expectations from AI, building systems that can be higher performance but also bring you that performance per [total cost of ownership].”

Stojanovic spoke with theCUBE’s John Furrier at theCUBE + NYSE Wired: AI Factories – Data Centers of the Future event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed scaling up AI systems and Ayar Labs’ approach to optical connectivity.

Ayar Labs scales up on AI workloads

There has been a huge build-out to facilitate AI and a corresponding increase of interest in technologies that avoid the heating issues and bandwidth limitations of traditional chip racks. With Optical I/O, Ayar Labs can stitch together thousands of GPUs into a fabric and take on compute-intensive AI workloads, Stojanovic explained.

“We allow both compute sockets and switches to have a much higher radix of connectivity,” he said. “That allows you to connect to more peers as well as have a flatter switch network. With a lot of bandwidth and a lot of radix, you also allow path diversity through a bunch of parallel switches and that significantly cuts down this end-to-end latency.”

Ayar Labs is focused on scaling up, as well as scaling out. Unlike electrical I/O, Optical I/O can keep up with the increasing number of GPUs without being beholden to limitations of cooling or density. This allows Ayar Labs to create a cluster-wide AI inference, saving both time and money, Stojanovic noted.

“In terms of density, we’re looking at essentially a decade of scaling ahead of us,” he said. “More importantly with partners like Alchip, we’re getting these optical engines into very complex AI socket packages. You have a CoWoS interposer there with a lot of memory and compute. Someone has to put all of these things together.”

Alchip’s partnership on co-packaged optics enables multi-rack AI clusters designed to eliminate data bottlenecks. True returns on AI investment depend on reimagining data-center operations — building a stronger hardware foundation to sustain next-generation performance, scalability and efficiency across increasingly dense, interconnected compute environments, Stojanovic emphasized.

“It’s the backend that’s very critical,” he said. “We’ve solved the frontend. The design works, the platform works. What you really need is this high-volume manufacturing backend that a tier one customer can access and essentially use to scale to extremely high volumes that we see in these AI applications.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of theCUBE + NYSE Wired: AI Factories – Data Centers of the Future event:

Photo: SiliconANGLE