Inside Amazon’s Low Key Plan To Dominate AI

Amazon’s absence from?this year’s generative AI bonanza has been a bit puzzling. The company invented Alexa, intuiting people’s interest in speaking with computers, yet when OpenAI released ChatGPT it seemed to cede the territory.

But rather than sitting out the game, Amazon is just waiting to play on its terms. Instead of building one AI product, it wants a piece of all of them. And it’s not shy about its ambition.

“I wouldn't be at all surprised if just the AI part of our cloud computing business was larger than the rest of AWS combined in a couple years,” Amazon VP Matt Wood told me in an interview at the Amazon Web Services Summit last week.

Rather than releasing just one product or large language model by itself, Amazon’s wants to enable companies building with generative AI to create any product using any model. Put another way, instead of developing one ChatGPT or GPT-4, Amazon wants to empower every would-be ChatGPT developer to use any GPT-like model and get going. Amazon will supply the model access, customization, and raw computing power to developers, and make its money as they build.

My full interview with Amazon VP Matt Wood, live from the AWS Summit, is available on Big Technology Podcast as of this morning:

You can listen on Apple Podcasts, Spotify, or your podcast app of choice.

“It is a great business opportunity,” said Andrew Lipsman , principal analyst at Insider Intelligence. “It’s smart strategically to focus on where the profits are.”

At the core of Amazon’s effort is a new product called Bedrock. Available inside AWS, Bedrock lets developers select from a range of AI models, including from Anthropic, AI21 Labs, and Stability AI. Using these models, developers can build their own products, like AI chatbots, and then run them on AWS’s infrastructure.?

Bloomberg, for instance, built BloombergGPT, a bot for financial information, on a Bedrock precursor called Sagemaker. To do it, the company took four decades of unstructured financial data and analytics, loaded it into AWS, added some other training material, and tuned the model. Bedrock should make such a process faster, with pre-loaded models in a catalog. Once built, Amazon then support the products. When people chat with BloombergGPT, for instance, it uses Amazon’s storage to work, so Amazon gets paid every step of the way.

“We get paid by providing compute capacity to actually do the model training, and for providing the access to the large amounts of storage that are needed,” said Wood. “You may train a model once a month, once a week, but you're going to be running predictions and inference and chatting with that model hundreds, thousands, tens of thousands of times a day.”

Competitors including Microsoft and Google offer similar capabilities, but Amazon has a few advantages. Without its own consumer chatbot, or a multi-billion dollar attachment to an AI research house, it’s pitching itself as a company with more neutrality and pragmatism than its peers. This could be compelling for developers looking for more customizability or assurances their data stays at home, a pressing issue for many. It’s also helpful for Amazon that so many internet companies already have their data on its cloud.?“You'd be surprised how many customers have exabytes of data on AWS,” said Wood. An exabyte is one billion gigabytes.

Amazon does have its own model, called Titan, that it offers alongside its menu of others. So it’s not entirely neutral. The company also develops its own AI-specific chips, which underly some of the computing, but it doesn’t sell them like NVIDIA does. Both efforts are meant to enhance the core service offering.

Long ago, Amazon learned there’s value to being first. It established its cloud services lead early and still dominates. But it’s playing catch up today as Microsoft appears to be in the lead, with?11,000 customers using its generative AI service?via a partnership with OpenAI.

In this case though, Amazon is joining right after the starting gun, with a plan that could work no matter what model or product wins. “We are three steps into a marathon race,” said Wood, “And I don't think anybody without a smile on their face could call a winner three steps into a marathon, right?”