Artificial intelligence just got more interesting—and a lot more personal. Amazon has unveiled its latest lineup of advanced AI systems along with something that could completely alter how businesses build their own models. But here’s where it gets controversial: can Amazon’s new approach really challenge AI giants like OpenAI and Google, or is it just playing catch-up in an already crowded field?
At its re:Invent conference in Las Vegas, Amazon introduced the second generation of its Nova AI models. The stars of the show are Nova Lite and Nova Pro, two large language models designed to compete with OpenAI’s GPT and Google’s Gemini families. There's also Nova Sonic, a real-time voice model, and Nova Omni, an experimental system that reasons through text, sound, images, and video. Though these models aren’t yet as famous as their competitors, Amazon’s promise lies in one differentiator—customization. The tech giant plans to make these models highly adaptable for each customer, a move that could resonate with its massive base of AWS users.
Building AI from the ground up
The most striking part of Amazon’s announcement is Nova Forge, a new platform that lets clients build what Amazon calls “frontier models.” Essentially, this tool allows users to insert their own data into the Nova 2 Lite and Pro models at multiple stages of development—even during the early pretraining phase that’s typically reserved for huge AI labs. Fine-tuning existing models like GPT or Gemini is already possible, but what Amazon is offering goes deeper—altering the base model itself to make it smarter in a company’s unique domain.
Rohit Prasad, head of Amazon’s AI division, explained that ever since Nova Forge was conceived, its goal has been empowerment. “Everyone wants an AI expert in their own field,” he said, describing this as a new, more open philosophy of training. The tool was originally developed for internal teams, including those improving Alexa and creating advanced AI agents.
Real-world tests: Reddit leads the way
One of the first organizations to try Nova Forge is Reddit. The social platform used the tool to craft a specialized system for flagging rule-breaking posts and comments. Chris Slowe, Reddit’s CTO, explained why traditional fine-tuning wasn’t enough—most AI models outright refuse to process controversial or violent material, even when that’s exactly the kind of data moderators need to review. With Nova Forge, Reddit combined custom pretraining and fine-tuning to build what Slowe proudly calls “a Reddit expert model.” This model not only understands how Reddit works but captures the nuanced language of its communities, making it extremely capable for tasks like content moderation. He added that this tailored model could power other future automation tasks as well.
Other prominent companies testing Nova Forge include Booking.com, Sony, and the biotech group Nimbus Therapeutics. Their participation highlights growing corporate interest in AI systems that go beyond typical, all-purpose models—especially ones that can deeply understand a company’s culture, data, or technical field.
Why this matters for the AI economy
Amazon’s timing might be strategic. A recent Bain & Company survey found that roughly three-quarters of U.S. firms view AI as a top business priority. Still, many struggle with limited know-how, lack of resources, or the sheer cost of building custom models from scratch. Nova Forge could simplify those challenges by offering enterprises a halfway point—unique, controllable models at a fraction of the traditional cost. Building a large AI model from zero can cost hundreds of millions of dollars; Amazon claims its solution is far cheaper, though it hasn't shared concrete figures.
That said, there’s a catch: Nova Forge is entirely tied to AWS. Customers get flexibility, but only within Amazon’s cloud ecosystem. Will that limitation put off companies seeking independence from big tech infrastructure?
Amazon’s quiet climb in the AI race
Although Amazon joined the large-scale AI race later than competitors, it has been quietly expanding its AI footprint. The company has embedded generative AI in its shopping experiences, most visibly through Rufus, a conversational shopping assistant. It’s also pumping billions into AI infrastructure and partnerships—like its major investment in Anthropic, the company behind the Claude language models. Anthropic’s latest systems even run on Amazon’s in-house Trainium processors, signaling Amazon’s ambition to compete with Nvidia’s hardware empire.
Amazon claims its Nova 2 Pro model holds its own—or even surpasses—top-tier systems such as OpenAI’s GPT-5, Google’s Gemini 3.0 Pro, and Anthropic’s Sonnet 4.5 on several benchmarks. Nova 2 Lite, the smaller sibling, reportedly performs on par with Claude 4.5 Haiku and GPT-5 Mini. Nova 2 Omni, meanwhile, dives into uncharted AI territory: a fully multimodal reasoning model that can interpret and generate across text, images, sound, and video. According to Prasad, no other mainstream company has publicly released anything quite like it.
The big question: innovation or entrapment?
Reddit’s Slowe believes one thing separates Nova from today’s other systems—pure adaptability. “I think it has a lot of potential,” he said. “For a broad range of use cases, it’ll outperform anything off the shelf.” Still, the question remains: is Amazon’s approach truly open innovation, or just a new way to bind customers more tightly to its cloud? As AI becomes foundational to every business, this debate could shape the future of who controls the next generation of machine intelligence.
What do you think? Does Amazon’s Nova Forge mark a breakthrough for customizable AI—or is it another walled garden in disguise? Share your thoughts and join the discussion below.
