AWS expands AI infrastructure, debuts agentic tools at re:Invent
AWS is entering a phase where companies will move beyond copilots and assistants to AI agents capable of completing full tasks across IT, operations, security, customer workflows and data-heavy environments, said CEO Matt Garman.
- Dec 3, 2025,
- Updated Dec 3, 2025 12:57 PM IST
Amazon Web Services (AWS) has placed a major strategic bet on agentic AI, with Chief Executive Officer Matt Garman unveiling new foundation models, upgraded AI infrastructure and autonomous enterprise -- AI tools at the re:Invent 2025 keynote aimed at reasserting the company’s position in the fast-moving cloud market.
Setting the context for the transition, Garman said, “Two years ago, people were building AI applications. Now, people are building applications that have AI in them.” He said AWS is entering a phase where companies will move beyond copilots and assistants to AI agents capable of completing full tasks across IT, operations, security, customer workflows and data-heavy environments.
“The next 80% to 90% of enterprise AI value will come from agents. This shift is going to have as much impact on your business as the internet or the cloud itself," he said, emphatic about the scale of the change.
“The time for simple copilots is over,” Garman said. “We are moving into the Age of the AI Agent, where AI turns from a technical marvel into something that delivers real, measurable business returns.”
The company introduced new models under the Amazon Nova family, including multimodal systems that process text, image, audio and video, and announced Nova Forge, a framework for enterprises to customise and deploy AI agents securely on AWS. Demonstrating the rationale, Garman asked, “What if you could integrate your data at the right time during the training of a frontier model, and then create a proprietary model that is just for you?”
The keynote also featured a major hardware push. AWS unveiled the next generation of AI-optimised servers powered by the Trainium 3 chip, developed alongside Nvidia technologies, promising faster training speeds and lower cost per inference for large-scale workloads. The new silicon targets enterprises building agents that run autonomously across large datasets and complex operational flows. Emphasising AWS’s position, Garman said, “AWS has the broadest AI infrastructure in the world.”
Garman said agentic AI will become the dominant layer of enterprise automation, adding that the demand for secure, predictable infrastructure will determine which cloud platform leads the next decade. He described AWS’s approach as balancing innovation with operational safety, noting, “We refuse what I call the ‘tyranny of the or’. You can’t pick A or B — we have to do both.”
AWS positioned these releases as a direct response to the rapid acceleration of AI adoption and intensifying competition from Microsoft and Google. The company also showcased new agentic-AI tools designed to automate software development, operations and workflow orchestration. “These agents work as an extension of your software development team,” Garman said.
Calling the product cycle a significant milestone, he added, “This marks a big leap forward in the journey towards unlocking the value of AI.”
The announcements reflect AWS’s intent to anchor AI development in enterprise-grade environments, with an emphasis on security, compliance and scalability. Analysts at the event noted that the combination of new models, agentic-AI frameworks and hardware upgrades marks one of AWS’s most aggressive product cycles in recent years, with the company signalling that the next phase of cloud growth will be driven by autonomous systems rather than general-purpose assistants.
“The long-term trajectory for enterprise AI is already visible-agents will ultimately recede into invisible infrastructure. With Nova models, AI Factories and Trainium silicon, AWS is laying the groundwork for ‘smart services’ where the orchestration and complexity of agents are fully abstracted away from the user experience. AWS is converging models, silicon and deployment architecture into an integrated agentic AI stack,” said Prabhu Ram, Vice President, Industry Research Group (IRG), CyberMedia Research (CMR).
Amazon Web Services (AWS) has placed a major strategic bet on agentic AI, with Chief Executive Officer Matt Garman unveiling new foundation models, upgraded AI infrastructure and autonomous enterprise -- AI tools at the re:Invent 2025 keynote aimed at reasserting the company’s position in the fast-moving cloud market.
Setting the context for the transition, Garman said, “Two years ago, people were building AI applications. Now, people are building applications that have AI in them.” He said AWS is entering a phase where companies will move beyond copilots and assistants to AI agents capable of completing full tasks across IT, operations, security, customer workflows and data-heavy environments.
“The next 80% to 90% of enterprise AI value will come from agents. This shift is going to have as much impact on your business as the internet or the cloud itself," he said, emphatic about the scale of the change.
“The time for simple copilots is over,” Garman said. “We are moving into the Age of the AI Agent, where AI turns from a technical marvel into something that delivers real, measurable business returns.”
The company introduced new models under the Amazon Nova family, including multimodal systems that process text, image, audio and video, and announced Nova Forge, a framework for enterprises to customise and deploy AI agents securely on AWS. Demonstrating the rationale, Garman asked, “What if you could integrate your data at the right time during the training of a frontier model, and then create a proprietary model that is just for you?”
The keynote also featured a major hardware push. AWS unveiled the next generation of AI-optimised servers powered by the Trainium 3 chip, developed alongside Nvidia technologies, promising faster training speeds and lower cost per inference for large-scale workloads. The new silicon targets enterprises building agents that run autonomously across large datasets and complex operational flows. Emphasising AWS’s position, Garman said, “AWS has the broadest AI infrastructure in the world.”
Garman said agentic AI will become the dominant layer of enterprise automation, adding that the demand for secure, predictable infrastructure will determine which cloud platform leads the next decade. He described AWS’s approach as balancing innovation with operational safety, noting, “We refuse what I call the ‘tyranny of the or’. You can’t pick A or B — we have to do both.”
AWS positioned these releases as a direct response to the rapid acceleration of AI adoption and intensifying competition from Microsoft and Google. The company also showcased new agentic-AI tools designed to automate software development, operations and workflow orchestration. “These agents work as an extension of your software development team,” Garman said.
Calling the product cycle a significant milestone, he added, “This marks a big leap forward in the journey towards unlocking the value of AI.”
The announcements reflect AWS’s intent to anchor AI development in enterprise-grade environments, with an emphasis on security, compliance and scalability. Analysts at the event noted that the combination of new models, agentic-AI frameworks and hardware upgrades marks one of AWS’s most aggressive product cycles in recent years, with the company signalling that the next phase of cloud growth will be driven by autonomous systems rather than general-purpose assistants.
“The long-term trajectory for enterprise AI is already visible-agents will ultimately recede into invisible infrastructure. With Nova models, AI Factories and Trainium silicon, AWS is laying the groundwork for ‘smart services’ where the orchestration and complexity of agents are fully abstracted away from the user experience. AWS is converging models, silicon and deployment architecture into an integrated agentic AI stack,” said Prabhu Ram, Vice President, Industry Research Group (IRG), CyberMedia Research (CMR).