Amazon Web Services’ re:invent has been in Vegas this week, so let’s take a look at some of the big announcements.
Key releases include a new AI training chip called Trainium3, which delivers up to 4x performance improvements for AI training and inference while reducing energy use by 40%. AWS also unveiled new features in its AgentCore platform, enabling developers to set boundaries for AI agents and allowing agents to log and remember user interactions. Additionally, AWS launched three new AI agents, including the Kiro autonomous agent that can write code and operate independently. The company also introduced four new models in its Nova AI family and a service called Nova Forge for accessing customizable AI models.
According to the announcement, these new models are designed to enhance machine learning capabilities, enabling businesses to tailor AI solutions to their unique needs.
Amazon Web Services has launched a new service, Amazon Nova Forge, designed to help businesses build their own AI models by combining proprietary data with existing Nova model frameworks. This effort aims to make AI model development more accessible, as creating these models from scratch can be very costly for many companies. According to CEO Matt Garman, Nova Forge offers access to multiple training checkpoints, allowing users to incorporate their data at various points during training. This service not only reduces technical and financial challenges but also produces customized models better suited to specific business needs. Once created, these models can be deployed on Amazon Bedrock, boosting security and scalability within the AWS ecosystem. This strategy highlights AWS’s commitment to flexibility and customer choice in AI development.
AWS has introduced new features for its AI agent platform, Amazon Bedrock AgentCore, designed to make building and monitoring AI agents easier for businesses. Major updates include a Policy feature that allows users to set boundaries for agent interactions using natural language, enhancing access controls to internal data and third-party applications like Salesforce and Slack. Additionally, AWS has launched AgentCore Evaluations, a collection of 13 pre-built systems for tracking AI agent performance, and AgentCore Memory, which allows agents to log user preferences over time.
AWS has introduced the AWS AI Factory, which offers dedicated AI infrastructure in customers’ data centers. This service includes Nvidia GPUs, Trainium chips, and AWS networking and storage features, enabling organizations to scale AI projects while maintaining compliance and data sovereignty. The infrastructure functions similarly to a private AWS Region and provides access to AWS managed services, including foundation models. Customers can integrate Nvidia hardware and software, and the AI Factory builds on AWS’s earlier Project Rainier designed for Anthropic.
Amazon is taking a strong position by suggesting that traditional artificial intelligence benchmarks might not be as important as previously believed. This reflects a larger trend in the tech industry, where companies are increasingly prioritizing practical AI applications over performance metrics alone. Supporting this view, Amazon’s approach highlights real-world usefulness rather than just theoretical performance, implying that the success of AI systems should be measured by their ability to solve actual problems rather than merely excelling on standardized tests. Industry experts point out that this shift could foster a more practical perspective on AI development, emphasizing user experience and application usefulness over benchmark scores.
According to a study by Omdia, expert partners within the AWS Partner Network can generate as much as $7.13 in services revenue for every $1 spent on AWS technology, with 82% of partners now delivering AI solutions. The recent AWS re:Invent event highlighted new agentic AI partner categories and the expansion of the AWS Marketplace, which aims to streamline procurement and support scalable adoption of AI technologies. Notably, AWS has increased its support for AI-specialized partners, introducing marketing development funds and reducing application processing time by up to 70%.
Why do we care?
Here’s what all this AWS noise really means: they’re trying to lock in the entire AI stack — from chips to agents to private regions — and they want you to build your services on top of it. That’s not inherently a problem, but let’s not kid ourselves. Every one of these features gets stickier the moment you adopt it. If you use their model tuning, their agent policies, their private-region infrastructure… you’re in deep. And AWS wants it that way.
What matters for MSPs is the shift in expectations. Customers are about to start asking, “How do we control what the agent can and can’t do? How is this logged? How do we audit it? Where does the model actually run?” You need answers. If you’re still talking about AI in terms of generic chatbots, you’re behind.
AI Factory is the loudest signal here: enterprise customers want sovereignty and compliance guarantees. And even if your SMB clients aren’t asking yet, they will soon — because vendors are normalizing it. You’ll need a governance story that scales down, not just a bunch of tools you stitch together.
The benchmark talk? That’s AWS trying to shift the conversation because they’re not winning the leaderboard game. For MSPs, that’s a reminder: evaluate vendors on what matters for your customers, not on whatever narrative the vendor is selling this quarter.
And that partner revenue multiplier? Great if you build standardized offerings. A trap if you slip into custom engagements that don’t scale.
Bottom line: This isn’t about the shiny announcements. It’s about taking control of your AI governance story before AWS — or anyone else — does it for you.