When the cloud community gathered in Las Vegas for AWS re:Invent one year ago, generative artificial intelligence was scarcely mentioned. Today, Amazon Web Services Inc. made a series of announcements during its annual conference that underscored how the technology has moved to the top of the cloud giant’s agenda.

Amazon Web Services Inc. unveiled what it called the “Generative AI Stack,” an approach designed to provide customers with applications for leveraging AI, new tools for building with large language models, and advances in infrastructure to accelerate model training and inference.

“The innovation around generative AI models is explosive,” said AWS Chief Executive Adam Selipsky (pictured). “It’s going to reinvent every application that we interact with at work and at home. We’re approaching the whole concept of generative AI in a fundamentally different way.”

Broad capabilities

Over the course of his nearly two-and-a-half-hour keynote appearance at re:Invent today, Selipsky provided specifics around a generative AI strategy that he previously hinted at in a wide-ranging exclusive pre-re:Invent interview with SiliconANGLE. The company’s focus centers around advances in silicon, building new generative AI capabilities on top of existing offerings, and continuing to pursue an innovative agenda through its key partners like Nvidia Corp. and Anthropic PBC.

“Having the broadest and deepest set of capabilities matters,” Selipsky said. “We set out to rethink IT infrastructure completely with AWS.”

A key ingredient in rethinking infrastructure will involve silicon. AWS has built its track record in silicon technology through the Nitro hypervisor and subsequent generations of Trainium, Graviton and Inferentia processors.

AWS put the role of silicon in context on Monday evening when Peter DeSantis, senior vice president of AWS Utility Computing, unveiled a set of advances for its serverless offerings and scale-out databases. “Nitro is the reason AWS got started building custom chips,” DeSantis said during his re:Invent presentation.

Today Selipsky announced AWS Trainium2, a purpose-built chip for generative AI and machine learning training, along with Graviton4. He also unveiled an expanded partnership with Nvidia that will include bringing the chipmaker’s DGX Cloud, an AI training-as-a-service platform, to AWS.

“DGX Cloud is Nvidia’s AI factory,” said Nvidia CEO Jensen Huang, who joined Selipsky on the re:Invent stage and noted that DGX Cloud will have 16,000 GPUs connected into one supercomputer. “It’s like 65 supercomputers in one.”

Potential rival to ChatGPT

This extra firepower in the silicon layer will ultimately drive a set of new offerings that AWS is counting on to attract enterprise interest and potentially provide an alternative to OpenAI’s ChatGPT generative AI solution. A key element of the cloud giant’s generative AI strategy was unveiled today with the announcement of Amazon Q, a generative AI-powered assistant for use in the workplace. The tool’s features will also be added to Amazon QuickSight, a cloud-native business intelligence service, allowing customers to craft tailor-made solutions for workplace applications, including call center management through Amazon Connect.

“You need this layer of capabilities that’s meant for servicing builders,” G2 Krishnamoorthy, vice president of AWS Analytics Services, Database and AI Leadership, said in an exclusive interview with SiliconANGLE on the application of Amazon Q to QuickSight in the generative AI stack. “It’s like an assistant that’s helps you with your tasks.”

Amazon Q will initially be deployed to developers and IT professionals, with the likelihood of broader use by businesses-at-large next year. Developers will be able to conversationally iterate new features into code within minutes.

“Amazon Q is also your business expert,” Selipsky said. “It’s completely customizable.”

That customizability using Amazon Q is a key differentiator for AWS as it seeks to deploy its generative AI stack. Customers want different models, according to Selipsky, who pointedly took note of OpenAI’s recent turmoil this month surrounding the abrupt departure and ultimate return of its CEO, Sam Altman.

“There is not going to be one model to rule them all,” Selipsky said. “You need to be trying out different models, you need a real choice of model providers. I think events of the last 10 days have made that very clear.”

Without directly calling out familiar names, Selipsky sprinkled his keynote remarks with references to AWS’ competitive position relative to “other cloud providers.” This was especially noticeable during his announcements on advances in purpose-built processors for generative AI and machine learning training.

“Meanwhile, a lot of cloud providers are just talking about their own machine learning chips,” Selipsky said.

Today’s announcements were preceded by a flurry of bulletins from AWS on Monday, with more than 80 news releases delivered at re:Invent before the first keynote speaker took the stage. The message was clear: AWS has no interest in relinquishing its position at the top of the public cloud food chain and it intends to press its advantage.

“At Amazon, we make bold bets,” Selipsky said. “The most important reinvention of all is yours.”

Photo: Robert Hof/SiliconANGLE