You might think Amazon’s biggest swing in the AI race was its $8 billion investment in Anthropic. But AWS has also been building in-house foundation models, new chips, massive data centers, and agents meant to keep enterprise customers locked inside its ecosystem. The company believes these offerings will give it an edge as businesses of all shapes and sizes deploy AI in the real world.

WIRED sat down with AWS CEO Matt Garman ahead of the company’s annual re:Invent conference in Las Vegas to discuss his AI vision, and how he plans to extend Amazon’s lead in the cloud market over its fast-rising competitors, Microsoft and Google.

Garman is betting that AI is a service that AWS can deliver more cheaply and reliably than its rivals. Through Bedrock, Amazon’s platform for building AI apps, he says customers can access a variety of AI foundation models while keeping the familiar data controls, security layers, and reliability that AWS is known for. If that pitch holds up, it could help AWS dominate in the AI era.

“Two years ago, people were building AI applications. Now, people are building applications that have AI in them,” said Garman, arguing that AI is becoming a feature inside large products rather than a standalone experiment. “That's the platform that we've built, and that's where I think you see AWS really start to take the lead.”

Many of the announcements at this year’s re:Invent fall along these lines. Amazon unveiled new, cost-efficient AI models in its Nova series; agents that can work autonomously on software development and cybersecurity tasks; as well as a fresh offering, Forge, that lets enterprises cheaply train AI models on their own data.

The stakes are high for AWS to get this right. While Amazon’s cloud unit dominated the smartphone era, smaller rivals like Google Cloud and Microsoft Azure have grown at higher rates since the arrival of ChatGPT. Microsoft and Google have surged by tightly integrating with frontier AI models—the technology underlying ChatGPT and Gemini, respectively—attracting enterprises eager to experiment with cutting-edge capabilities.

This rise of AWS’s rivals has raised questions about Amazon’s broader AI strategy, and how the incumbent will fare in the years to come.

Garman says he’s been hearing these concerns for years, but less so in recent months. He argues that the tide is turning, pointing to AWS’s stronger-than-expected results in the company’s third quarter as evidence that his strategy is working.

The counter argument to his approach comes from other AI leaders, who believe that AI is a more fundamental shift in computing and will force companies to completely rethink their approach to product development. In a world where AI is a true paradigm shift, cutting-edge AI capabilities may take precedence, and incumbents like AWS could be in a more precarious position.

AI Efficiencies

AI seems to be driving major organizational changes inside of Amazon. In October, the company announced it would lay off 14,000 people as it invests more in AI. Those layoffs came just a few months after Amazon CEO Andy Jassy said Amazon would need fewer people in certain areas because of AI.

Internally, Garman says that AI tools are significantly accelerating some engineering teams’ work as employees shift from writing code themselves to directing a team of AI agents. He says one AWS team recently did a big rewrite of an internal codebase, expecting that it would take 30 people roughly 18 months to complete. With the use of AI, Garman claims the team was able to complete the task with just six people in 71 days.

Some Amazon employees have a less optimistic tone when it comes to the company’s embrace of AI. In November, more than 1,000 anonymous Amazon employees signed a petition warning that the company’s “aggressive” AI rollout could come at a cost to the environment.

Garman notes that the upside of AI agents are not unlimited—though his focus here is around agent management, not environmental impact.

“Agents are most effective when you ask them to do things that … you actually know how to do yourself,” he said. “So these are not replacements for people. They are ways to make people more effective at their jobs.”

Still, Amazon says the efficiency gains of AI are also making their way to Amazon customers. Reddit was given early access to Amazon’s new Forge service, and was able to train an AI model on millions of content moderation decisions, according to AWS. Reddit said the AI model it made with Amazon developed a “social intuition,” and can now help with content moderation for millions of its online communities.

The AI Bubble

While Amazon is betting big on AI, Garman is skeptical of the exuberance sweeping the broader industry. This year, early stage labs like Mira Murati’s Thinking Machines Lab and Ilya Sutskever’s Safe Superintelligence nabbed multibillion-dollar investments, despite that they’re not yet deploying widely used products.

“When people talk about a bubble, I think those are the deals that are most at risk,” said Garman, referring to AI startups that have raised billion dollar seed rounds. “Where it's a $3 billion valuation for a startup with no lines of code. Maybe, but maybe not. Those are the ones where I think there's open questions.”

But Garman insists that Amazon’s AI investments are justified. The company brought on 3.8 gigawatts of new infrastructure capacity in the past 12 months, and just announced an up to $50 billion investment in AI data centers for the US government. Amazon says it’s still able to sell off new capacity as soon as it comes online.

“I see the value that so many companies are getting out of AI today, and I don't see that there's any pullback,” said Garman. “They're getting real returns. They're delivering real value to their customers. And so for us, that is a good signal, and we're still in the early stages of what that value is going to be.”

As Silicon Valley continues to chase superintelligence and AGI, AWS seems to be taking a more grounded approach. Whether that’s enough to remain king of the cloud providers remains to be seen.

Question for the audience:

With the launch of Gemini 3, I’m hearing a lot of folks in the AI industry say that AI progress hasn’t stalled, despite what the skeptics might think. I’d love to hear from Model Behavior readers: What are you doing with AI today that wasn’t possible 12 months ago? On the flip side, how is the technology the same (capability-wise) as it was last year? Respond in the comments below or email me—happy to keep things off record.

This is an edition oftheModel Behavior newsletter. Read previous newslettershere.