Skip to content

AWS re:Invent 2025 - Keynote with CEO Matt Garman

In this keynote, Matt Garman, CEO of AWS, focuses on how AWS is innovating across all aspects of cloud computing to empower customers and partners to build a better future (2:51). The central theme revolves around the explosion of invention with AI and the shift from AI assistants to AI agents, which are expected to deliver significant business returns and scale human impact by 10x (11:35-13:09).

Key takeaways from the video include:

AWS Growth and Scale (3:26-5:39)

AWS has grown into a $132 billion business, accelerating at 20% year over year. The company's infrastructure is vast, with S3 storing over 500 trillion objects and daily average requests of over 200 million per second. More than half of new CPU capacity comes from Graviton, and Amazon Bedrock powers AI inference for over 100,000 companies, with many processing over 1 trillion tokens each (28:58-29:13). AWS also has the largest and most broadly deployed AI cloud infrastructure globally, spanning 38 regions and 120 availability zones (4:52-5:21).

Customer-Centric Approach and Partner Ecosystem (5:40-7:27)

Everything at Amazon starts with the customer, and AWS aims to give developers the freedom to invent (10:19). AWS serves millions of customers across various industries, including financial services, healthcare, media, and government agencies. Security is a top priority, leading the U.S. Intelligence Community and companies like Nasdaq and Pfizer to choose AWS. The keynote highlights the crucial role of its massive network of partners (6:29), including SaaS providers and system integrators. A special affinity for startup customers is noted, with more unicorn startups built on AWS than anywhere else (6:52-7:00).

Focus on AI and Agentic Systems (11:35-13:31)

The keynote emphasizes the "explosion of invention with AI" and the rapid iteration of technology. Matt Garman believes that AI agents are an inflection point in AI's trajectory, moving it from a technical wonder to something that delivers real business value (12:30). He envisions billions of agents within every company and field (12:46), accelerating discoveries and scaling human impact.

AI Infrastructure Innovations (13:47-25:46)

AWS provides the most scalable and powerful AI infrastructure (13:53), emphasizing operational rigor and attention to detail that leads to industry-leading GPU reliability (15:00-15:28).

P6e-GB300 Instances (15:56)

Announcement: P6e-GB300 instances are now generally available, powered by Nvidia's latest GB300 NVL72 systems.

Takeaway: These instances provide best-in-class compute for demanding AI workloads, offering over 20x the compute compared to the previous P5en generation.

AWS AI Factories (17:43)

Announcement: AWS AI Factories are now generally available.

Takeaway: This enables customers to deploy dedicated AWS AI infrastructure in their own data centers for exclusive use, operating like a private AWS region while offering access to leading AWS AI infrastructure and services, including UltraServers, SageMaker, and Bedrock.

Amazon EC2 Trn3 Ultraservers (22:32)

Announcement: Amazon EC2 Trn3 Ultraservers are now generally available.

Takeaway: These are AWS's most advanced UltraServers, featuring the first three-nanometer AI chip (Trainium3) in the AWS Cloud. They offer the industry's best price-performance for large-scale AI training and inference, with 4.4x more compute and 3.9x memory bandwidth compared to Trainium2.

Amazon Bedrock and Model Choices (27:40-35:40)

Amazon Bedrock is presented as a comprehensive inference platform (27:50) for fast-tracking generative AI applications from prototype to production. AWS is committed to offering a broad selection of models in Bedrock, including open-weights and proprietary models, to ensure customers have the best options for their specific use cases (29:32-29:59).

Amazon Nova 2.0 Family (31:58)

Announcement: The new generation of Amazon's foundation models, Nova 2.0, is now generally available.

  • Nova 2 Lite: A fast and cost-effective reasoning model suitable for a broad set of workloads, excelling at instruction calling, tool calling, code generation, and information extraction
  • Nova 2 Pro: The most intelligent reasoning model for highly complex workloads, particularly strong in instruction following and agentic tool use
  • Nova Sonic 2.0 (32:29): A next-generation speech-to-speech model enabling real-time, human-like conversational AI with improved latency and expanded language support
  • Nova 2.0 Omni (35:20): The industry's first reasoning model that supports text, image, video, and audio input, and generates text and image output, providing unified multimodal reasoning

Amazon Nova Forge for Custom Models (41:59)

Announcement: Amazon Nova Forge is now generally available, introducing "open training models."

Takeaway: Integrating unique company data and IP deeply into models is crucial for unlocking huge value and differentiating businesses (38:05). Nova Forge allows customers to blend their proprietary data with Amazon-curated training datasets at every stage of model training, creating "Novellas" (proprietary models) that deeply understand their specific information without losing core foundational capabilities (41:44).

Agent Services and Tools

Amazon Bedrock AgentCore (1:05:55, 1:29:23)

Announcement: Policy in Amazon Bedrock AgentCore (Preview) and AgentCore Evaluations (Preview) are now available.

Takeaway: These are tools for deploying and operating highly capable agents securely at enterprise scale, showing strong momentum with over 2 million SDK downloads.

New Autonomous Agents

Announcements:

  • Kiro Autonomous Agent (Preview) (1:41:59)
  • AWS Security Agent (Preview) (1:47:23)
  • AWS DevOps Agent (Preview) (1:50:11)

Takeaway: New agent services to automate tasks in security and DevOps.

Additional Service Announcements

Compute Instances

Announcements:

Serverless and Storage

Announcements:

  • Lambda Durable Functions (GA) (1:59:53)
  • S3 Batch Operations Now 10x Faster (GA) (2:01:00)
  • Automatic replication for S3 Tables (GA) (2:01:37)
  • S3 Access Points for FSX for NetApp ONTAP (GA) (2:01:53)
  • Amazon EMR now with no storage provisioning required (GA) (2:03:29)

Security and Monitoring

Announcements:

  • GuardDuty Extended Threat Detection for ECS (GA) (2:03:56)
  • Unified Data Store in CloudWatch (GA) (2:05:22)

Database Services

Announcements:

  • Additional storage volume for Amazon RDS for SQL Server and RDS for Oracle (GA) (2:06:10)
  • Optimize your CPUs in Amazon RDS SQL Server (GA) (2:06:29)
  • Amazon RDS for SQL Server Developer Edition (GA) (2:06:46)
  • Database Savings Plans (GA) (2:07:16)

Resources