Attendees pass the Amazon Web Services logo during AWS re:Invent 2024, a conference hosted by Amazon Web Services at the Venetian Hotel in Las Vegas on December 3, 2024.
Noah Berger | Getty Images
Amazon has discovered a way for cloud clients to extensively customize the AI models they generate. The problem is, this system costs $100,000 a year.
Nova Forge, powered by Amazon Web Services, gives organizations access to Amazon’s AI models at various stages of training, allowing them to incorporate their own data early in the process.
Already, companies can now fine-tune large language models after training. Nova Forge results are highly dependent on data provided by customers. Nova Forge customers also have the option to refine open-weight models, but training data and computing infrastructure are not included.
Organizations that assemble their own models may end up spending hundreds of millions or billions of dollars, so using Nova Forge is more affordable, Amazon says.
AWS released its own model under the Nova brand in 2024, but these are not the first choice for most software developers. By the middle of this year, Amazon-backed Anthropic controlled 32% of the enterprise LLM market, followed by OpenAI with 25%, according to a July study by Menlo Ventures. google 20% and meta Amazon Nova’s share was less than 5%, according to a Menlo spokesperson.
Nova models, like Anthropic’s Claude 4.5 models, are available through AWS’s Bedrock service for running models on Amazon cloud infrastructure.
“We’re a frontier lab that’s been customer-focused,” Rohit Prasad, Amazon’s chief scientist for artificial general intelligence, said in an interview with CNBC. “Our customers wanted it. We invented it on their behalf to make this happen.”
Prasad said Nova Forge is also used by Amazon’s internal customers, including the company’s stores and the team responsible for its Alexa AI assistant.
reddit We needed an AI model to moderate sophisticated content on the many subjects people discuss on social networks. Engineers found that the Nova model, enriched with Reddit data through Forge, performed better than larger, commercially available models, Prasad said. Booking.com, Nimbus Therapeutics, Nomura Research Institute and Sony also use Forge to build models, according to Amazon.
Organizations can request Amazon engineers to help build Forge models, but that assistance is not included in the new service’s $100,000 annual fee.
AWS is also announcing new models for developers at its Reinvent conference in Las Vegas this week.
Amazon says the Nova 2 Pro is an inferential model that has been tested to perform at least as well as Anthropic’s Claude Sonnet 4.5, OpenAI’s GPT-5 and GPT-5.1, and Google’s Gemini 3.0 Pro Preview. Inference involves performing a series of calculations, which may take extra time depending on the request for a better answer. According to Prasad, Nova 2 Pro will be available in early access to AWS customers with Forge subscriptions. This means Forge customers and Amazon engineers can try out Nova 2 Pro at the same time.
Nova 2 Omni is another inference model that can process incoming images, audio, text, and video and generate images and text. Amazon says it’s the first inference model with such a wide range of capabilities. By providing multifaceted models, Amazon hopes to reduce the cost and complexity of incorporating AI models into applications.
Tens of thousands of organizations use the Nova model every week, Prasad said. AWS claims to have millions of customers. Prasad said Nova is Bedrock’s second most popular model family. The top group of models is from Anthropic.
WATCH: Amazon to launch AI conference next week: Maxim’s Tom Forte talks about what’s next
