“Make your own AI model”, AWS’s bet for the next stage of artificial intelligence

The presentation was made by the CEO of AWS, Matt Garman, at the opening talk of re:Invent 2025 in Las Vegas, Amazon’s annual conference to announce its new developments to society. The news covers models and services, in addition to chips and data houses tailored.

It is the largest event in the world techie, a festival of technical and commercial profiles that flood the main hotels in the Strip from Sin City to listen to talks, receive training, learn about success stories or generate new business. In an era where everything happens through the cloud, this is the place where the cloud comes down to earth for a while and starts networking.

Garman spoke on the stage set up in the main hall of The Venetian Hotel for just over two hours to tell everything the company did during 2025. More than 6,000 people listened live below.

“It was a more technical conference than in previous years, focused on solution developers. The cloud democratized access to tools for companies and AI is allowing anyone to develop software,” summarizes Rafael Mattje, AWS Solutions Architecture leader for the Southern Cone.

For the CEO, there will not be a single AI model that can monopolize the market or be useful to all customers. That is why they think about a future with a range of solutions for different needs. “A world with millions of AI agents,” he anticipated during his presentation.

One of the first announcements was the incorporation of 18 new models to the catalog of Amazon Bedrockthe generative AI platform that allows you to pivot between different models such as those of OpenAI, Meta or DeepSeek to obtain better results. Among those joining are Mistral AI, Gemma 3 (from Google), Nemotron (Nvidia) or GPT OSS Safeguard (OpenAI), among others.

Amazon also adds its own, like the Nova 2 Omni, which promises to receive text, images, videos and voice as inputs, and return both text and images. The objective is to facilitate creative processes.

One of the highlights was the launch of Amazon Nova Forge, a service for companies to create their own frontier models from scratch. It is the equivalent of having a ChatGPT trained from the beginning specifically to operate within the client’s business.. “It’s like learning a language: if you do it as a child it’s easier; learning as an adult is more complex,” Garman summarized.

The service allows you to create a model that deeply understands the organization’s own knowledge, while minimizing risks such as catastrophic forgetting (when adding new data causes essential information for the model to work) and preserving fundamental capabilities such as reasoning.

The novelty is that companies They can decide at what point in the construction of the model they provide their data. By adding them in the first pre-training instances, the new model learns from the beginning on a specialized basis for the type of organization.

AWS conducted a pilot test with six clients, including the social network Reddit, which trained its model to help moderate comments on the forum, known for its toxicity. “With less information it became more precise,” Garman summarized.

Other companies that participated in the pilot were Siemens and Sony, which trained its model with information from consumer feedback on many of the products it launches, such as video games. PlayStation 5 or movies.

Another important announcement was the launch of AWS AI Factories, a type of service designed for organizations that are wary of handing over their data—such as banks or governments— can develop and deploy AI applications at scale, but keep their data in physical spaces own instead of storing them in the cloud.

For AWS, whose business is based precisely on offering storage and processing in the cloud, it represents a paradigm shift: offer their services completely “turnkey”. “Creating a high-performance AI factory requires a complete set of management, database, storage and security services, a complexity that few clients are willing to assume,” the company explained.

The service includes the latest AI accelerators, such as chips NVIDIA AI Computing y Trainium next-generation, plus high-speed, low-latency networks from AWS, along with comprehensive AI services such as Amazon Bedrock or SageMaker.

These solutions are designed to help meet sovereignty and regulatory requirements by controlling where data is processed and stored. One of the current customers is Saudi Arabia, which is building an AI zone that will have about 150,000 dedicated chips.

There were also hardware announcements with the launch of the new self-developed AI processor, Trainium 3. According to AWS, it is approximately four times faster and 40% more efficient than the previous version.

These processors seek to compete with Nvidia GPUs, offering a cheaper alternative that also reduces Amazon’s dependence on that company’s chips.

A while before the central announcements, a joint project with the national government and six provinces was presented from the Latin American offices of AWS to promote Argentine startups through the use of cloud services.

This is the AWS Activate program, implemented in coordination with the Ministry of Productive Development of the Nation and the provincial governments of Córdoba, CABA, Entre Ríos, Jujuy, Mendoza and Tierra del Fuego. In total there will be a fund of 48 million dollars in credits to use AWS tools, technical support, training, management tools and advice to optimize costs and accelerate new product development.

Companies must be linked to digital services and be less than ten years old. They may receive up to $100,000 in credits. Since its launch, the program supported companies like Airbnb, Reddit, Stripe, and DoorDash in their early stages. To apply you must enter the website argentina.awsentrena.com, which was launched this Monday.

“This announcement represents our commitment to collaborating with governments and leveling the playing field for startup founders throughout Argentina,” he told Clarion Lorena Zicker, AWS leader for the Public Sector in Argentina, Chile, Paraguay and Uruguay.