With the race for artificial intelligence supremacy long under way, many of the larger industry players are already scrambling to create and drive cutting-edge AI innovations across a slew of use cases.

One of those is Amazon Web Services Inc., and its whirlwind of updates across storage, processing and large language models are drawing considerable attention.

“What people don’t realize is that [AWS] won the cheap storage war a while ago,” said Andy Thurai (pictured, left), vice president and principal analyst at Constellation Research Inc. “Now they’re making it faster — I mean, the claim if it holds true, 10 times faster and about half cheaper. Storage is becoming a major issue for all the AI/ML programs and by providing this, I think they’re going to win that. The cloud war is over, but the AI/ML stack war is still on.”

Thurai and Sarbjeet Johal (right), founder and principal of Stackpane, spoke with theCUBE industry analyst Dave Vellante at the “Supercloud 5: The Battle for AI Supremacy” event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed AWS’ multifaceted strategy, from custom silicon to AI infrastructure and a unique model playground approach.

Improvements across the storage and processing gamut

The key unveiling on storage is the new Amazon S3 Express One Zone storage class. It’s been designed to deliver up to 10x better performance than the standard class, all while handling requests in the hundreds of thousands per second and maintaining single-digit millisecond latency, according to Thurai.

Moving on to silicon, there’s been a move by the big players toward custom silicon tailored specifically for AI-related tasks. AWS, for its part, just took the wraps off the Trainium2 and Graviton4 processors. The new Trainium chip is custom-designed for training neural networks, with specific emphasis on LLMs and foundational models. The new Graviton entry is underpinned by Nvidia Corp. hardware and is designed to handle a broad range of cloud workloads, with a promised 30% improvement in compute performance and 75% more memory bandwidth.

“I think there’s a compression there of the experience because the whole industry is learning,” Johal said. “Microsoft has learned from it as well, and there’s a lack of experience that will play into it, so they will gain. They’re not like eight years behind or five years behind, let’s say. They started at five years, but [now] they’re maybe two years behind.”

AWS’ messaging and company direction have made it clear that its strategy is not just about having the best silicon, but providing a comprehensive platform for AI. Unlike some of its competitors, AWS is not striving to create a one-size-fits-all LLM, but rather emphasizes giving customers the flexibility to choose, train and fine-tune their models according to their specific needs.

“They want to become the model playground for everything,” Thurai explained. “They’ll let you build the models, build apps using that and figure out what fits you — and then help you fine-tune that model. There’s a low-code, no-code way to fine-tune it, and then you can have your own inference. That’s totally the opposite story of what the other guys are telling.”

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