We are entrenched firmly in the generative artificial intelligence era — it is the differentiating factor for today’s enterprise and business landscapes.

The industry is seeing larger players such as Amazon Web Services Inc. double down on harnessing gen AI’s transformative power, with a palpable dedication to integrating the tech across a gamut of services.

“We are super excited about all the things we launched as well because AWS’ strategy with generative AI is really resonating with customers in a big way,” said Swami Sivasubramanian (pictured), vice president of database, analytics and machine learning at AWS. “We are already seeing customers in every industry, from healthcare with Pfizer to fintech with Intuit. They’re already innovating, but also not just at the models level but even at the infrastructure level to application level.”

Sivasubramanian spoke with theCUBE industry analysts Dave Vellante and 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 increasing importance of data, AI and builders, creating a beautiful symbiosis that propels technology into the next era of intelligence augmentation.

The scale of infusion: From LLMs to HyperPod

The magnitude of gen AI infusion at AWS is vast, particularly because the totality of that current state isn’t recent — rather, it’s a culmination of over two decades of experience in large language models, according to Sivasubramanian.

Crucial for the company has been its integration and innovation at every layer of the stack, from infrastructure collaboration with NVIDIA to advancements in the software stack with SageMaker and the revolutionary HyperPod, reducing training time by up to 40%. HyperPod removes the undifferentiated heavy lifting involved in building and optimizing ML infrastructure for training foundation models.

“We are innovating in every layer of the stack, but at the infrastructure layer, we have been investing with our strategic collaboration with NVIDIA, which we expanded with our GPU-based instances to custom silicon with Trainium and Inferentia,” Sivasubramanian explained. ” [There’s also] all the innovations we are doing in the software stack with SageMaker, with things like HyperPod.”

With advances in high-value gen AI areas such as image generation comes complexity. For instance, image LLM training runs now cover much larger data arrays and, as such, take more time to complete, according to Sivasubramanian. Given this increased complexity, there’s been a need for concomitant innovation in areas such as constant checkpointing, better failure detection and automatic distribution, all integrated into the HyperPod framework, he added.

“Building a gen AI app is a lot more than just taking an LLM and plugging it through,” Sivasubramanian explained. “You still need to worry about things like contextual relevance, you need to worry about guardrails, you need to pick the right tool for the right job. So that’s where we saw all the features like Bedrock model evaluation, guardrails and embeddings, and vector databases that we launched play a big role.”

Moving on, there needs to be a critical assessment of the role of data engineers, chief data officers and network engineers in the upcoming wave of generative AI. Admittedly, the next focus would be on abstracting data and AI as code for developers, reshaping the data engineering landscape, Sivasubramanian concluded.

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