This Tuesday, AWS CEO Matt Garman led the company's 14th annual re:Invent keynote, an event that made clear that AWS is betting big on AI in the enterprise, and specifically, on agents that can learn and work autonomously as the future of AI. A dizzying amount of news and announcements focused on AI infrastructure and new products supported the message that AWS is working hard to become a leading provider of enterprise grade hardware infrastructure, models and ready-to-use agents.

The centerpiece hardware announcement was Trainium 3, AWS's first 3-nanometer AI chip, now generally available. Trainium 3 Ultra Servers reportedly deliver 4.4x more compute and five times more AI tokens per megawatt compared to their predecessors, with individual instances combining up to 144 chips for 362 FP8 petaflops of performance. AWS also previewed Trainium 4, promising six times the compute power.

On the software front, AWS launched Nova 2, its next-generation foundation model family. Nova 2 Pro leads with frontier-level reasoning and agentic capabilities, outperforming GPT-5.1, Gemini 3 Pro, and Claude 4.5 Sonnet on key benchmarks. Nova Forge, a model customization service, gives customers exclusive access to Nova training checkpoints so they can combine their proprietary data with Amazon's curated datasets during pre-training to create custom, domain-specific "Novella" models that deeply understand proprietary knowledge bases and IP while retaining Nova core reasoning capabilities.

Perhaps most significant were three new Frontier Agents (all currently in preview): Kiro for autonomous software development, AWS Security Agent for proactive vulnerability scanning, and AWS DevOps Agent for incident resolution. These long-running, scalable agents represent AWS's vision for AI-driven automation across the entire development lifecycle. Notably, AWS intends to jumpstart Kiro adoption by launching a program in which early-stage startups can receive a year's worth of credits for Kiro Pro+ for up to 100 users. Interested parties can sign up here, although it should be noted that restrictions apply.

Beyond AI, AWS rolled out substantial updates to core services. The company introduced Database Savings Plans, offering up to 35% savings across all database services for customers who enter into a 1-year consistent hourly commitment. Additional announcements included expanded RDS storage capacity to 256 terabytes for SQL Server and Oracle, new EC2 instance families, and S3's maximum object size increasing tenfold to 50 terabytes.

To get the complete scoop on these announcements and many others, the full keynote can be revisited in the AWS Events YouTube channel.