Low prices. Wide selection. Tough-to-beat convenience.
Those have been crucial keys to Amazon’s e-commerce dominance for the past two decades.
Now the tech titan seems to be doubling down on this winning playbook by borrowing pieces of it to propel a new AI strategy from its Amazon Web Services division that is focused in part on low prices and wide selection.
To be clear, it’ll be a long time before the business world can judge the effectiveness and financial sustainability of Amazon’s approach. But, if successful, the game plan would go a long way toward both quieting critics who argue Amazon is playing catch-up in the AI wars, while also future-proofing the company’s standing as one of the world’s most powerful and influential technology corporations for decades to come.
Amazon executives have been unveiling crucial pieces of their AI strategy at their flagship AWS Re:Invent conference in Las Vegas this week. One key element is a new portfolio of in-house-built foundation models, known as FMs or LLMs, dubbed Nova, that can handle text, image, and video queries, respectively.
The introduction of a new class of Amazon’s own AI models could, on the surface, be confounding since the company has already invested $8 billion into Anthropic, maker of the popular Claude family of gen AI models. But, as my colleague Sharon Goldman recently noted, Amazon is betting that there is “never going to be one tool”—or AI model—to rule them all. In short, Amazon believes that enterprises will want choice in models, whether from Amazon, Anthropic, or other tech giants like Meta.
Amazon actually started pushing this idea of offering a selection of AI models through a single API to business customers when its AWS division introduced a service called Amazon Bedrock last year. Through Bedrock, business customers could choose from a relatively limited selection of AI models—but a selection nonetheless—to train for their own needs and to serve as a foundation for their own gen AI applications.
On Wednesday, Amazon doubled down on the strategy by announcing Bedrock Marketplace, which offers a total of 100 AI models. The LLMs in the marketplace come from a host of different companies, with some designed for specialized purposes.
“Finding and evaluating these models can be challenging and costly,” Amazon said in its blog post announcing the marketplace. “You need to discover them across different services, build abstractions to use them in your applications, and create complex security and governance layers. Amazon Bedrock Marketplace addresses these challenges by providing a single interface to access both specialized and general-purpose [foundation models].”
Looking back to Amazon’s e-commerce business, a core piece is the Amazon Marketplace, where hundreds of thousands of outside merchants list products for sale that make up 60% of all the goods sold through Amazon. Amazon complements the selection of these marketplace sales by also selling its own inventory of goods, including some under its own brand names when a certain product category or price point isn’t being filled by the marketplace sellers or Amazon’s brand partners.
Similarly, Amazon is offering businesses an enterprise AI version of the marketplace that one could imagine will only expand in selection in the future. (It’s also worth noting that Amazon’s core AWS business offers a marketplace of more than 10,000 software tools covering categories from cybersecurity to data analytics.)
Low prices have also been another hallmark of Amazon’s retail business. Amazon aggressively matches prices from other retailers, and throws two giant discount events that attract heavy spending and new Prime customers. (The FTC has argued in its antitrust suit against Amazon that the e-commerce giant artificially inflates consumer prices around the web by penalizing merchants who sell products for less at other retailers, but that’s a topic for another day.)
And sure enough, Amazon CEO Andy Jassy’s slide deck unveiling the new Nova AI models started with price: “75% more cost effective,“ was the first feature called out.
Simon Willison, an independent AI researcher, ran a quick test and agreed, writing on the social network app Bluesky that Amazon’s “price and performance [are] competitive with the Google Gemini family, which means they are _really_ inexpensive.”
“With this release I think Amazon may have earned a spot among the top tier of model providers,” Willison added. “Maybe we need a new FAANG acronym that covers OpenAI, Anthropic, Google, Meta, and Amazon. I like GAMOA.”
Amazon leaders won’t care what the new acronym is called as long as they earn a place in it. If they do, the company’s longtime hallmarks of low prices and selection will likely be key reasons why.
Are you a current or former Amazon or AWS employee with thoughts on this topic or a tip to share? Contact Jason Del Rey at jason.delrey@fortune.com, jasondelrey@protonmail.com, or through messaging apps Signal and WhatsApp at 917-655-4267. You can also message him on LinkedIn or at @delrey on X.
