AWS CEO Matt Garman declares AI agents bigger than internet as Amazon stakes claim in AI race

AWS CEO Matt Garman Positions Amazon at AI Frontier

AWS CEO Matt Garman used the company's annual re:Invent conference in Las Vegas to make a bold declaration: AI agents will be more transformative than the internet or cloud computing itself3

3

. "The world is not slowing down," Garman told attendees, emphasizing that "the advent of AI agents has brought us to an inflection point in AI's trajectory"3

3

. His comments signal Amazon's intent to reassert its cloud dominance as Microsoft and Google have grown faster since ChatGPT's arrival.

The stakes are high for AWS in the AI race. While Amazon's cloud unit remains the market leader, it saw 20% growth last quarter compared to 34% for Google Cloud and 40% for Microsoft's Azure4

4

. Matt Garman acknowledged past concerns but told Axios that "a year ago, there were questions about whether we'd missed the wave, but now, most people are building their production systems in AWS"4

4

. He pointed to stronger-than-expected third-quarter results as evidence his strategy is working.

Bedrock Platform Reaches Multi-Billion Dollar Scale

Garman revealed that Bedrock, Amazon's managed service for building AI apps, has become a multi-billion dollar business with over 100,000 customers2

2

5

5

. This marks the first public disclosure of the Bedrock platform's scale in these terms. AWS has doubled the number of AI foundation models available on Bedrock in just one year, adding 18 new open-weight models from Google, Minimax, Mistral, NVIDIA, and OpenAI5

5

.

The company unveiled its Nova series of cost-efficient AI models at re:Invent 2025. Nova Light handles cost-efficient reasoning, Nova Pro tackles complex reasoning across documents and video, and Nova Sonic adds multilingual speech-to-speech capabilities5

5

. A unified Nova multimodal model processes text, images, video, and speech as inputs, solving what Garman described as a real need for creative teams working across different formats.

Custom Chips and AI Infrastructure Push

AWS is betting heavily on custom chips to compete on cost effectiveness with NVIDIA-dominated infrastructure. The company has already deployed more than 1 million Trainium chips and unveiled Trainium3 UltraServers with significant performance and energy gains5

5

. Garman previewed Trainium4, which delivers "over 6x the FP, 4x performance, 4x more memory family and 2x more high bandwidth memory capacity" while doubling power efficiency5

5

.

The AWS CEO emphasized the company's 15-year partnership with NVIDIA, noting that "NVIDIA itself trains its largest systems on AWS"5

5

. AWS also introduced "AI factories" that let the company place its AI infrastructure, software, and governance controls inside customer environments to meet sovereignty and policy requirements.

AI Agents Transform Enterprise AI Strategy

Garman's vision centers on autonomous agents that can handle complex, long-running tasks without constant human oversight. AWS unveiled new autonomous agents for software development, cybersecurity, and DevOps at re:Invent1

1

. Bedrock Agent Core brings building blocks like memory, gateway, identity, policy controls, and real-time evaluations that monitor agent behavior and raise alerts when quality slips5

5

.

The shift to AI agents has fundamentally changed Garman's thinking about Amazon's future needs. Speaking at re:Invent, he revealed that six or seven years ago, he calculated Amazon would need to hire a million developers to deliver on its product roadmap2

2

. With AI, he no longer believes that's necessary. Massive technical projects that once required "dozens, if not hundreds" of people might now be delivered by teams of five or 10, thanks to AI and agents2

2

.

Competing Against Microsoft, Google, and Anthropic

AWS faces intense competition from Microsoft's OpenAI partnership and Google's Gemini integration, which have attracted enterprises eager to experiment with cutting-edge capabilities1

1

. Amazon's $8 billion investment in Anthropic gives it access to Claude models, but the company lacks a flagship frontier model directly comparable to GPT-5 or Gemini 3 Pro4

4

.

Garman told Axios that AWS is "increasingly the cloud where customers are putting real production workloads due to its combination of capabilities and cost effectiveness"4

4

. He argues that AWS's strength lies in offering customers choice across multiple AI foundation models while maintaining familiar data controls, security layers, and reliability. "Two years ago, people were building AI applications. Now, people are building applications that have AI in them," Garman said, arguing that AI is becoming a feature inside large products rather than standalone experiments1

1

.

Garman was candid about Amazon's weaknesses, admitting that "one of the things that Amazon is particularly bad at is being a fast follower. When we try to copy someone, we're just bad at it"2

2

. Instead, AWS focuses on thinking from first principles about solving customer problems where it has differentiated expertise. The company's multi-layered approach builds core infrastructure like compute and storage at the bottom, offers databases and AI models in the middle alongside partners, and builds applications selectively at the top2

2

.