AWS Unveils Graviton5 Chips, Nova AI Models, and Frontier Agents at re:Invent 2025

AWS reveals Graviton5 processors, an expanded Amazon Nova model family, and autonomous frontier agents at its annual re:Invent conference. These innovations focus on maximizing cloud performance, enabling open AI training, and extending software development capabilities.

AWS unveils a massive lineup of technological innovations at re:Invent 2025, headlined by the introduction of Graviton5, the company's most powerful and efficient custom CPU to date. The new Graviton5-based Amazon EC2 M9g instances deliver up to 25% higher performance than the previous generation, featuring 192 cores per chip and a cache that is five times larger. Over half of all new CPU capacity added to AWS this year relies on Graviton processors, with major enterprise clients like Adobe, Airbnb, and SAP already leveraging these chips for superior price performance.

The tech giant also significantly expands its Amazon Nova portfolio with four new models designed for reasoning, multimodal processing, code generation, and agentic tasks. A standout addition is Nova Forge, which pioneers an "open training" approach by giving organizations access to pre-trained model checkpoints so they can blend proprietary data with Amazon-curated datasets. Additionally, the Nova Act model achieves a breakthrough 90% reliability rate for browser-based user interface automation, helping companies like Hertz accelerate their development velocity by five times.

Rounding out the major announcements is the debut of frontier agents, a new class of artificial intelligence tools that function as an autonomous extension of software development teams. These scalable agents work continuously for hours or even days without requiring human intervention. Alongside these AI and processor advancements, AWS highlights the availability of Trainium3 UltraServers and the launch of AWS AI Factories, which allow customers to implement robust AI infrastructure directly within their existing on-premises data centers.

Read More at the original source →