Amazon’s bona fides in the space are well established, having been a player in artificial intelligence and machine learning long before the ChatGPTs and Bards of the world were publicly released. Former Fortune editor Brian Dumaine wrote a book in 2020 about how Amazon founder Jeff Bezos realized early on that imbuing machine learning into every facet of the company would allow it to gather data to constantly improve itself. 

Much as it did with Amazon Web Services, which practically birthed the cloud computing industry that now powers the internet’s biggest companies, including its competitors, Amazon’s A.I. strategy is focused on cementing its position as a major player across the entirety of the A.I. supply chain. 

“Every single business unit inside of Amazon is working intensely and very broadly on generative A.I.,” Jassy says.  

Jassy shed some light on Amazon’s A.I. game plan, outlining three macro layers: the computing capabilities, the underlying models, and what Jassy refers to as the “application layer,” for example, ChatGPT or Bard.  

How Amazon will compete in A.I.

New chips strong enough to generate the vast amounts of computing power needed for the nascent technology will become key manufacturing parts in the future. Currently chipmaker Nvidia has about 83% market share, making the market ripe for a new entrant with Amazon’s technical expertise and deep pockets. So far, AWS has developed two different chips: Trainium, for training machine learning models, and Inferentia, which powers the inferences that ultimately yield a given output. Both have better price-performance ratios than other chips on the market, according to Jassy, which he says is key given how much computing power will be needed to power A.I. moving forward. 

These and other chips Amazon hopes to develop will all be used to power the “foundational models” upon which all generative A.I. applications are built. Jassy says he expects that as a result only about six to eight of these models will be the basis of almost all generative A.I. tools moving forward. But as of now those models are prohibitively expensive—to the tune of billions of dollars—and will require years to perfect, Jassy says, making them inaccessible to ordinary developers, aspiring A.I. startup founders, and even established companies. To address that problem Amazon created Bedrock, a service that sells large machine learning models as a service to customers who don’t want or can’t develop their own. 

“What [customers] really want is to take that foundational model and customize it with their own data, without leaking any of that custom data into the generalized model, and they want it with all the same platform and security capabilities they get in AWS,” Jassy says of Bedrock.  

It is to machine learning what AWS is to server space. If successful, Bedrock could become a go-to service for any company with ambitions to develop their own generative A.I. application. 

Amazon doesn’t plan to sit out that part of the A.I. race, either. Jassy says as of now, Amazon is focused on building a generative A.I. tool that can help developers write code more quickly and in figuring out the right applications for the technology in improving the customer experience. He does, however, acknowledge the “overwhelming majority” of those applications will be built by other companies—and obviously he hopes they’ll use AWS’s suite of A.I.-specific tools to do so.

Amazon has long been bullish about its chances in the A.I. arms race. Last month, AWS CEO Adam Selipsky poured cold water on the current furor surrounding the most recent applications, saying companies were only three steps into a 10K. 

“You ask yourself the question—where are the different runners three steps into a 10K race?” Selipsky says. “Does it really matter? The point is, you’re three steps in, and it’s a 10K race.”