AWS CEO Matt Garman talks to CRN about its new Trainium3 AI accelerator chips being the ‘best inference platform in the world,’ AI openness being a market differentiator versus competitors, and ‘supercharging’ its agentic AI product Kiro.
Amazon Web Services CEO Matt Garman says the company’s newly launched Trainium3 AI accelerator chips are the “best inference platform in the world,” while its AI openness strategy is helping to win enterprise customer market share.
“We make it so if you want to pull in Gemini models from Google, you can use those with our AgentCore,” said Garman in an interview with CRN.
“We think that innovation happens in lots of different places, not just inside the walls of Amazon. We want our customers to be able to benefit from that ecosystem of innovation,” said Garman. “It is a difference in philosophies of how we operate, and our competitors don’t always operate that same way.”
[Related: AWS’ Nova 2 AI Models Launched At re:Invent 2025 As CEO Touts New Innovation ]
AWS re:Invent 2025
Thousands of AWS customers, developers and partners flocked to Las Vegas this week to attend AWS re:Invent 2025.
The Seattle-based $132 billion cloud giant launched a slew of innovation at the conference, including the general availably of its new Trainium3 chips for AI training and interfering.
“Trainium3 is actually going to be the best inference platform in the world,” Garman told CRN.
AWS sees Bedrock with Trainium3 as the world’s leading inference engine and a business that will one day possibly be as large as Amazon EC2. Garman and his team expect Trainium3 to be a major hit with enterprise customers as it will provide better price-performance options.
“[Trainium3] will be the most efficient, most effective, the best cost performance, the lowest latency and the best throughput,” Garman said.
In an interview with CRN, Garman takes a deep dive into AWS’ new Trainium3 AI chips, “supercharging” its agentic AI development offering Kiro, and why AWS’ AI openness strategy will win versus competitors like Microsoft and Google Cloud.
What’s one big launch you’re excited about for re:Invent 2025?
The biggest one that we have at re:Invent is really about Trainium3.
Trainium2 is the one that’s out on the market. Now we’ve seen some enormous traction from Trainium2, particularly from our partners at Anthropic who we’ve announced Project Rainier, where there’s over 500,000 Trainium2 chips helping them build the next generations of models for Claude. That’s been a remarkable success.
We’re very excited to make Trainium3 [generally available] at re:Invent.
It gives you a 4X compute boost over Trainium2. It gives you much better latency and better throughput, much lower cost for serving models.
Trainium is actually quite good for inference.
Talk about AI inferencing when it comes to Tranium3.
Trainium3 is actually going to be the best inference platform in the world.
It will be the most efficient, most effective, the best cost performance, the lowest latency and the best throughput.
So super excited about that as an inference platform.
Today, Trainium2 already serves more than half of the tokens through Bedrock on Trainium2 under the covers.
So when you go to Bedrock and you just ask for the model, you never see the compute. So you don’t really know that it’s more than half of the tokens that go through Bedrock already are served by Trainium2.
I’m excited that Trainium3 should really blow that number out too, just because it’s such a fantastic platform.
Partners say AWS’ openness strategy in terms of things like AI model choice and AI model selection has helped AWS partners differentiate themselves versus competitors Google Cloud and Microsoft. Your thoughts?
It is a difference in philosophies of how we operate, and our competitors don’t always operate that same way.
I think they want to own everything, which is OK, they can go on that strategy. For us, we love the partner model for a number of reasons.
We think that innovation happens in lots of different places, not just inside the walls of Amazon. We want our customers to be able to benefit from that ecosystem of innovation.
So if you think about AI as an example, or even just think about agent development, we didn’t build AgentCore as a closed system where you have to use everything inside of AgentCore and not use anything else. We built it so that you can pull the very best of pieces.
We think that AWS is going to build some, and has built some, pretty fantastic building blocks to help you build scalable, secure, high-quality agents.
But we make it so if you want to pull in Gemini models from Google, you can use those with our AgentCore.
If you want to use frameworks from Amazon, like Strands, awesome. If you want to use frameworks from LlamaIndex and open-source frameworks, no problem—pull those in.
You can really mix and match and pull all these different pieces in.
Does this openness help AWS and partners with enterprise customers?
As an enterprise, you can decide, ‘What do I want to use inside of Bedrock, where I can have all this security and everything put together? And where is there really differentiated value that I can pull from outside?’
We see this across AWS.
We often have people that have their EC2 Auto Scaling system set up, and their S3 system set up, and Snowflake as their analytics engine. That has been a successful recipe for us for a very long time.
You can keep going across the board: They use CrowdStrike for their security, they use MongoDB for their document repository, etc.
We have this partner ecosystem that gives us a really rich set of opportunities, and then we build capabilities so that those partners can more deeply integrate their solutions into AWS.
We keep building more and more capabilities so that third-party offerings look more like AWS.
It’s easier for customers to consume. They can use their same permissioning.
They can click without having to do a bunch of configuration.
We’re trying to keep lowering that bar so that something like an offering from CrowdStrike can look like it’s a native AWS solution, so that customers can just easily use those.
For us, it’s built on AWS. So our business grows when their business grows and it’s good for customers.
What are some of the AI launches you’re excited about at re:Invent?
Kiro, which is our software developer IDE [integrated development environment] and CLI [command-line interface]has really pioneered this agentic development mode where development teams are now building with agents.
We are supercharging the Kiro experience with some agents that are going to be able to completely go and autonomously do software development for customers.
In addition to that, we’ll have autonomous agents that can also do operations and DevOps and do traditional SRE [site reliability engineering] work and CloudOps for users to both: either handle issues and/or proactively look for issues in your environment and suggest improvements.
We’ll have a security agent that can proactively look at your code, make recommendations on security improvements you can do, and actually do automated pentesting before you release it out into development.
So our goal there is these agents that can really supercharge a development team.
How will AWS’ new AI agents ‘supercharge a development team’?
Before, a small group of developers were often overwhelmed trying to do the blocking and tackling of delivering for customers.
Our goal is we can 10X a development team to go and build something great for customers.
Our view is that a team probably has 100 ideas that they just haven’t gotten to yet because they have a relatively small development team, and they’re trying to get going. But they would love to be able to iterate much faster and make their product better for their end customers.
We want to launch some of these agents with the idea that it can take that small development team and help them 5X or 10X their output so that they can more quickly get to the broad set of ideas that they have.
Then once they start to get success—because through the Marketplace they’re able to reach this global audience—they’re able to scale doing operations, scale doing security, scale doing all of those pieces without necessarily having to hire hundreds and hundreds of people to keep up with that growth.
So that’s one area that I’m super excited about.