AWS re:Invent 2025: Takeaways from Matt Garman’s keynote

AWS re:Invent 2025 kicked off with an opening keynote from AWS CEO Matt Garman. He explained how AWS is strengthening their foundational building blocks while continuing to innovate for AI agents.

Here are the main announcements and takeaways from his keynote, including new services to support AI agents, cost efficiency enhancements, and security updates. Plus, learn more about what makes Pluralsight the AWS Emerging Business Applications Partner of the Year.

Before Matt’s official opening keynote address, AWS announced the 2025 Geo and Global AWS Partner Awards at re:Invent. These awards recognize organizations who have modeled specialization, innovation, and cooperation over the past year. A panel of AWS experts selected the winners and finalists based on strict criteria with a third-party audit.

We’re excited to share that Pluralsight has been recognized as the Emerging Business Applications Partner of the Year. This award recognizes rising AWS Technology Partners with a horizontal business area focused around Business Applications. 

Our recognition as an AWS award winner underscores our position as a trusted tech skill development leader. It also acknowledges the role we play in helping AWS customers drive innovation and build solutions on Amazon Web Services. 

Read more about the Global AWS Partner Awards.

Matt acknowledged the changes AI has brought and the fact that many organizations still haven’t seen the financial returns generative AI promised. 

But Matt thinks AI agents will change that. “Agents will have as much impact on your business as the internet or the cloud,” he said.

To get value from AI agents, you need a few things:

• AI infrastructure

• An inference platform

• Your data 

• Tools to build agents

Most of Matt’s key product announcements fell into one of these categories, underscoring AWS’s emphasis on AI agents for future transformation.

New Amazon EC2 P6e-GB300 UltraServers are now generally available. These instances provide 1.5x GPU memory and 1.5x FP4 compute (without sparsity) compared to P6e-GB200.

These new EC2 P6 instances are ideal for intensive applications that require high context or use inference techniques like reasoning and agentic AI.

AWS AI Factories is AI infrastructure that AWS customers can deploy in their own data centers. They combine critical AI infrastructure, like Trainium chips, storage and databases, security, and high-speed low-latency networking, with AI services like Bedrock and SageMaker AI.

AWS AI Factories save time and make it easier for organizations to deploy large-scale AI applications while meeting data sovereignty and compliance requirements. This makes them especially useful for public sector organizations and regulated industries.

AWS Trainium is their custom AI chip. The latest version, AWS Trainium3, powers Trn3 UltraServers, delivering 4.4x more compute, 3.9x the memory bandwidth, and 5x more tokens per megawatt compared to Trn2 UltraServers.

Amazon EC2 Trn3 UltraServers will enable faster, lower-cost generative AI training, including things like reinforcement learning, Mixture-of-Experts (MoE), reasoning, and long-context architectures.

But AWS isn’t stopping there: They’re already working on AWS Trainium4, which will further accelerate AI training.

Choosing the right model is critical, and AWS announced 18 new, fully managed open weight models for Amazon Bedrock. This includes models from:

• Google

• MiniMax AI

• Mistral AI

• Moonshot AI

• NVIDIA

• OpenAI

• Qwen

More models means more options to find the best fit for your use case. In Amazon Bedrock, you can try out, and even switch between, models without needing to rewrite code or change infrastructure.

Learn more about each model's use cases.

Amazon Nova announced new Nova 2 models. There are a few versions: 

• Nova 2 Lite: Ideal for everyday workloads, it can process text, images, and videos to generate text. Use it for things like customer service chatbots, document processing, and business automation.

• Nova 2 Pro: As Amazon’s most intelligent reasoning model, it’s great for highly complex tasks like agentic coding, long-range planning, and sophisticated problem-solving.

• Nova 2 Sonic: This speech-to-speech model enables human-like conversational AI. This makes it great for customer service applications, AI assistants, and interactive voice experiences.

• Nova 2 Omni: The first of its kind, this multimodal reasoning and generation model can process text, images, video, and speech while generating text and images.

There are currently two ways to build a custom model:

1. Build a custom model from scratch, or

2. Use open weight models as a base and then modify it with fine-tuning, reinforcement learning, and other techniques

The first option is expensive and impractical for most organizations. The second often causes models to forget foundational training when introduced to new concepts. 

Amazon Nova Forge empowers organizations to bypass these issues and build their own frontier models. They can use a mix of their own datasets and curated training data to develop custom models with proprietary or industry-specific data that actually understand their niche.

Nova Act helps “developers build, deploy, and manage fleets of reliable AI agents for automating production UI workflows.” 

This service allows developers to build agents with natural language prompts and then deploy them to complete tasks like: 

• Updating data in a customer relationship management (CRM) system

• Coordinating shipments

• Submitting health insurance claims

• Writing test scenarios

Before you deploy your AI agents to customers or even employees, you need to make sure you have the right guardrails in place.

New features in Amazon Bedrock AgentCore allow you to define real-time controls and checks to ensure you deploy trustworthy agents.

• Policy in AgentCore: Control how agents interact with tools and data. This feature intercepts AgentCore Gateway tool calls before they run using policies with fine-grained permissions.

• AgentCore Evaluations: Continuously evaluate agent behavior and quality to catch issues before they impact customers.

Building applications requires a complex interplay between multiple services like payment processing and AI agent orchestration. Traditionally, this meant developers needed to integrate multiple infrastructures, handle failures, and more.

Now, developers can use AWS Lambda durable functions to build multi-step applications and AI workflows without needing to write custom code or manage even more infrastructure. 

Durable functions automatically “checkpoint progress, track progress, automatically retry on failures, and suspend execution for up to one year at defined points, without paying for idle compute during waits.”

MLflow within Amazon SageMaker AI allows organizations to experiment with AI and evaluate the performance of AI models, applications, and agents. 

MLflow now scales automatically, ramping up for demanding tasks and then scaling back when idle. This serverless capability eliminates the need for capacity planning and allows developers to spend more time on actual experimentation.

Matt closed the keynote with a rapid-fire list of new features, including:

• AWS Security Agent: This new agent proactively secures your applications throughout the development lifecycle to prevent vulnerabilities early on.

• AWS Security Hub: Identify and respond to critical security vulnerabilities faster with this unified security operations hub. 

• GuardDuty Extended Threat Detection for ECS and E2: Amazon GuardDuty now includes coverage for your containers and EC2 instances.

• Amazon S3 Vectors: The first cloud object storage with native support for storing and querying vector data is now generally available.

• Amazon Relational Database Service (RDS) capabilities: New features improve cost and storage optimization for RDS. This includes SQL Server Developer Edition support, expanded storage capabilities for RDS for Oracle and RDS for SQL Server, and CPU optimization options for RDS for SQL Server.

• GPU acceleration and auto-optimization for Amazon OpenSearch Service: Build vector databases faster and more cost-efficiently with these enhancements.

• New and enhanced AWS Support plans: New support plans bring AI into the equation for more proactive problem-solving.

• Unified Amazon CloudWatch experience: The new CloudWatch capabilities bring together data management and analytics for improved operations, security, and compliance.

• Database Savings Plans for AWS Databases: This new savings plan can help customers save up to 35% across all supported database services.

AWS is going all-in on AI agents—do you have the skills to keep up? Build AI and AWS skills with Pluralsight’s hands-on skill development platform.

Learn more about Pluralsight’s partnership with AWS and explore Pluralsight on the AWS Marketplace.