AWS re:Invent: AWS Forges
a new AI future

Amazon Web Services has unveiled a new platform called Amazon Nova Forge, in one of the most significant AI announcements of the year. The move reframes the entire contest for AI leadership, allowing AWS to advance along a path untouched by rivals like Google and OpenAI. Instead of competing for attention with ever-larger public models, AWS is giving enterprises the ability to create high-end AI systems built entirely from their own data and operated at full industrial scale.

“What if you could integrate your data at the right time of training during the training of a frontier model, and then create a proprietary model that was just for you?” asked AWS CEO Matt Garman during his opening keynote address at the annual AWS re:Invent conference in Las Vegas last week.

The question positioned Forge as a response to a frustration that has grown steadily across global enterprises. Large organisations produce vast reserves of information, yet seldom gain full value from it through generic AI tools.

“Almost every customer I talk to wishes that they could somehow teach the model to really understand their data, really understand their deep domain knowledge,” said Garman. He described corporate information as a strategic resource that carries years of operational insight, industry nuance and institutional memory, yet often remains siloed or underutilised. Forge addresses this gap by allowing companies to feed their own knowledge into a high-end model during the training process itself, rather than bolting it on at the edges.

Zeus Kerravala, principal analyst at US firm ZK Research, told Gadget that Amazon Nova Forge was a significant pivot by AWS that effectively charted a new course in the AI race.

“This positions them not just as a provider of foundational models, but as a platform for custom model creation,” he said. “This strategy allows AWS to move beyond direct competition with the likes of Google and OpenAI at the top of the model stack – where they risk being out-innovated – and instead, focus on democratising the ability to build proprietary, high-performance, industry-specific models.

“By giving customers, the tools to fine-tune Nova models from an early checkpoint, blend them with their own unique data, and access AWS-curated data sets, they are solving the critical problem faced by enterprises: creating highly effective agents that rely on proprietary context and taxonomy, without the massive expense and time required to train a frontier model from scratch.”

He said the approach significantly strengthened AWS’s competitive positioning by leveraging its core strengths: “customer trust, massive scale, and a leadership position in infrastructure sovereignty with the ability to simplify the complex”.

Rather than competing on the quality of a single general-purpose model, AWS is competing on the quality of the entire development and deployment lifecycle for the most sophisticated enterprise applications.

This, said Kerravala, would ensure their platform remains indispensable as the “backbone of AI” for most of the market. Prior to Garman taking the reins at AWS last year, it had been a product company. Now, “it has become the most complete AI platform vendor as they can address all aspects”.

Along with Forge, Garman also unveiled a fresh AWS approach to agentic AI – systems that make decisions without prompting. He announced Frontier Agents, which operate across development, security and operational tasks that typically consume considerable human time, and run for extended periods.

Their origins come from experiments inside Amazon, where internal teams tested early versions of autonomous coding assistance.  The experiments succeeded, and from this emerged a new idea: agents that handle sustained, multi-stage work across systems, repositories and workflows.

Frontier Agents include specialised assistants for security and development, carrying a significant portion of the technical workload inside many large organisations.

Garman also addressed a more traditional reality that affects businesses across every sector: the weight of legacy technology. Enterprises face layers of ageing infrastructure, old frameworks, outdated databases and abandoned platforms that demand constant upkeep.

“One of the biggest pain points today for development teams is technical debt,” he said. “Seventy percent of IT budgets today are consumed by maintaining legacy systems. We built AWS Transform to help customers move away from their legacy platforms.”

Transform Custom expands this further by allowing enterprises to create agents tailored to their own systems. This lifts a considerable barrier for companies with internal languages, long-retired systems or specialised workflows.

Veteran analyst John Furrier, CEO of SiliconANGLE, told Gadget that Amazon had in effect found a route to the front of the AI pack.

“Amazon has an installed base, they have developers and builders, and they’ve got enterprise customers,” he said. “Enterprise customers want value out of their data as fast as possible, so getting some wins and showing real value keeps them in Amazon’s camp. The builders just want to build new, fresh things, and if it’s easier and they get more value, they’ll stay with Amazon. Amazon is employing the Microsoft strategy: take care of your customers and don’t make them switch.

“Microsoft was way behind in the cloud game. The reason why they’re still in it and growing is they’ve convinced their customers that their cloud is good. It wasn’t at the beginning, and then it got better and better. They bought time, and now they have market share. Amazon is in a similar situation with AI. They’re behind, and now they’re going to use their customer base and their loyalty with entrepreneurs and builders and developers to maintain value.”

*Arthur Goldstuck is CEO of World Wide Worx, editor-in-chief of Gadget.co.za, and author of The Hitchhiker’s Guide to AI – The African Edge.