Amazon’s Bold Generative AI Race with Custom AWS Chips: Setting the Pace or Catching Up?

Infotrading.io - From the heart of Austin, Texas, an unassuming building hides the future of AI. It’s here that Amazon is quietly channeling its energies to design microchips set to revolutionize generative AI. The chips, dubbed Inferentia and Trainium, present a tempting choice for those weary of the escalating costs of Nvidia GPUs.

Amazon Web Services CEO, Adam Selipsky, was notably confident in an interview with CNBC, asserting, “The entire world would like more chips for doing generative AI… and we’re poised better than anyone on Earth to meet this demand.” It’s a hefty claim, especially when giants like Microsoft and Google have already taken monumental strides in the generative AI race. Microsoft, for instance, caught eyeballs by hosting the phenomenal ChatGPT and committing a staggering $13 billion to OpenAI.

Meanwhile, not one to be outdone, Google stepped into the limelight with Bard, its own large language model, and a weighty investment in Anthropic, OpenAI’s contender.

Yet, when Amazon unveiled its generative AI model, Titan, and the supporting Bedrock service, many couldn’t help but wonder: Was Amazon merely joining the race, or was it setting a pace of its own? Chirag Dekate from Gartner weighs in, “Amazon usually pioneers markets. But this time, they seem to be playing catch-up.”

But here’s where things get spicy. Amazon’s custom silicon may just be the dark horse. These unique chips could potentially provide the edge Amazon needs in the generative AI race. Dekate drops the truth bomb: “Microsoft doesn’t have a Trainium or Inferentia.” That’s a zinger!

Let's roll back the tape a bit. Amazon's tryst with custom silicon isn’t new. Its journey started in 2013 with Nitro, AWS's most-used chip. Fast forward, and Amazon's Graviton emerged in 2018, positioning itself as a rival to behemoths like AMD and Intel.

The CNBC team got an exclusive sneak-peek of the Austin chip lab, where Amazon's star chips, Trainium and Inferentia, are birthed and honed. Matt Wood, VP of product, provided insights: “Trainium offers about a 50% boost in price performance relative to any other AWS machine learning model training method.”

Amazon’s generative AI offerings are diverse. AWS HealthScribe, for instance, aids doctors in drafting patient visit summaries. Not to mention, CodeWhisperer has reportedly supercharged developer productivity by 57%. Add to that, AWS’s recent commitment of $100 million towards a generative AI innovation hub.

But it’s not all smooth sailing. The rapid expansion of AI has sparked security anxieties. Many companies fret about proprietary information being fed into the training data of large public models. Addressing this, Selipsky comments, “Any model accessed via our Bedrock service operates in an isolated virtual private cloud environment, fully encrypted with AWS's gold-standard access controls.”

Closing thoughts? Amazon might be a latecomer to the generative AI party, but it’s brought the fireworks. Whether it's playing catch-up or redefining the race, one thing's for sure: Amazon is all in on the generative AI race.