/dq/media/agency_attachments/UPxQAOdkwhCk8EYzqyvs.png)
Follow Us/dq/media/media_files/stEsbnVba8AJz69pC0cL.jpg)
Amazon Web Services (AWS) announced an expansion across its artificial intelligence portfolio, introducing new foundation models, a unique "open training" platform, dedicated software development agents, and next-generation, high-performance training hardware. The combined announcements target three main industry challenges: the cost of custom model development, the complexity of reliable workflow automation, and the escalating expense of frontier model training.
Amazon Nova: Models, Open Training, and Reliable Agents
Amazon expanded its Nova portfolio with four new models, aiming to establish industry leadership in price-performance across various tasks, including reasoning, code generation, and multimodal processing. The company introduced four variants: Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic, and the unified multimodal model, Nova 2 Omni.
Pioneering Open Training with Nova Forge
AWS introduced Nova Forge, a service pioneering "open training." This approach grants organizations access to pre-trained, mid-trained, and post-trained Nova model checkpoints. Customers can blend their proprietary data with Amazon’s curated datasets at every stage of the model training process to create highly optimized custom models, which Amazon refers to as "Novellas."
Customization Power: Organizations typically face compromises when trying to integrate deep proprietary knowledge into AI models. Nova Forge solves this by allowing customization that goes beyond simple fine-tuning, deeply integrating an organization’s expertise into the model's core knowledge.
Customer Example: Companies like Reddit use Nova Forge to replace multiple specialized models with a single, more accurate solution, improving content moderation efficiency.
Nova Act for Reliable UI Automation
The company also introduced Nova Act, a new service for building highly reliable AI agents for browser-based UI automation workflows. Nova Act achieves a breakthrough 90% reliability for workflows like updating customer records or submitting insurance claims.
Training Method: Nova Act achieves high reliability by training a custom Nova 2 Lite model usingreinforcement learning across thousands of simulated web environments ("web gyms"). This approach ensures the agent excels at practical, UI-based tasks.
Customer Example: Hertz accelerated its development velocity by five times using Nova Act, demonstrating its ability to handle repetitive, browser-based tasks quickly and accurately.
Frontier Agents: Autonomous Software Development Teams
AWS unveiled a new class of frontier agents designed to act as autonomous, scalable extensions of a software development team, capable of working for hours or days without constant human intervention. These agents focus on three crucial areas of the software development lifecycle:
| Agent Name | Core Function | Customer Impact |
| Kiro Autonomous Agent | Acts as avirtual developer, maintaining context across sessions and learning from code reviews and feedback to handle bug triage or code coverage improvements across multiple repositories. | Focuses human developers on strategic priorities by handling routine development work independently. |
| AWS Security Agent | Acts as a virtual security engineer, proactively reviewing design documents and scanning pull requests against organizational security requirements and vulnerabilities. | SmugMug used the Security Agent to catch a complex business logic bug that traditional, existing security tools could not detect. |
| AWS DevOps Agent | Acts as an on-call operational team, instantly responding to issues, correlating data across observability tools (like CloudWatch, Dynatrace), and pinpointing root causes to reduce mean time to resolution. | Commonwealth Bank of Australia uses the agent to build faster, more resilient banking infrastructure. |
Trainium3 UltraServers: Lowering the Cost of Frontier AI
Addressing the spiraling infrastructure cost of training increasingly complex AI models, AWS announced the general availability of Amazon EC2 Trn3 UltraServers. These systems are powered by AWS’s Trainium3 chip, built on the 3nm process technology.
Infrastructure Power: The Trn3 UltraServer system packs up to 144 Trainium3 chips into a single integrated unit.
Performance Metrics: The new infrastructure delivers up to 4.4 times more compute performance and 4 times greater energy efficiency compared to the previous Trainium2 UltraServers. Customers using Trn3 achieve 3 times higher throughput per chip and 4 times faster response times for inference.
Cost and Time Savings: This advancement helps organizations reduce training times from months to weeks and cut training and inference costs by up to 50% compared to alternatives.
Customer Example:Decart is using Trainium3 for real-time generative video, achieving 4 times faster inference at half the cost of competing GPU solutions. Anthropic, Karakuri, and Ricoh are also early adopters reducing their training and inference costs.
These releases establish a vertically integrated AI stack for Amazon, from the underlying Trainium3 silicon to the foundational Nova models and the application-level Frontier Agents.