It’s no secret that there is a surge of spending coming on generative artificial intelligence in this ongoing experimental phase, which represents a big conversation at this year’s AWS re:Invent conference. Infrastructure is key, and there are supply chain problems when it comes to graphics processing units.

Still, when it comes to generative AI, many things have not changed. It’s important to really take stock of the infrastructure layer, according to Prasad Kalyanaraman (pictured), vice president of infrastructure services at Amazon Web Services Inc.

“The reason why it’s important is because this is not something you can just enable overnight,” he said. “This is over years of innovation that we’ve actually done.”

Kalyanaraman spoke with theCUBE industry analyst John Furrier at the “Supercloud 5: The Battle for AI Supremacy” event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed the innovation that has taken place and what it takes to run generative AI workloads.

On innovation

When it comes to innovation at AWS, the most obvious may be chip design, according to Kalyanaraman. The company has been working from Graviton to Inferentia to Trainium.

“Over the course of time, we keep innovating on the chip design, and we keep innovating on the price performance of these chips as well,” he said. “But then there’s a lot of other things in the infrastructure that’s important.”

Security is job number one, according to Kalyanaraman. That remains true in the generative AI space, and the company believes its advancements with its Nitro System works in its favor in that regard.

“Our innovations on making sure that we have secure communications between our data centers, and we encrypt the traffic, all of that is actually fundamental to generative AI. It’s also important that customers’ data are secure,” he said.

It’s also important to innovate on the networking side of things and on the supply chain, according to Kalyanaraman. On the latter subject, AWS builds and designs its own servers.

“We’ve gone down to the details of trying to make sure that the hardware infrastructure is [efficient] from a power and from a sustainability perspective,” he said. 

When it comes to running any kind of generative AI workload, whether it’s training or inference, it’s not just about putting the chips in there, even if one figured out a way of connecting them. It’s also important to actually figure out where one’s data will be stored, according to Kalyanaraman.

“You need a really highly scalable data source, which is performing. And so you look at services like S3, and you look at Zero-ETL,” he said. “You need the data layer that comes with it.”

Beyond the data layer, if an organization is running it just for an application, it can probably put a certain level of perimeter security on top of it. But it’s important to ensure security at every layer of the infrastructure stack, according to Kalyanaraman.

“All the way from the Nitro layer, where we actually build it on the servers, to the network layer, and then to the perimeter as well. So, that’s the second part,” he said. “You need the data services to provide you the data, then you need the network.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the “Supercloud 5: The Battle for AI Supremacy” event:

Photo: SiliconANGLE