In the BotHub feed on Habr, this shift was captured very precisely: AWS is no longer just “a cloud where you rent a ready-made model.” It is increasingly becoming a platform where companies assemble their own LLMs tailored to their industry, processes, and internal data — without handing over control to the outside. The trigger was the presentation of the Amazon Nova 2 family and the Nova Forge service, which allows intervention in the model at a much deeper level than traditional fine-tuning.
The Nova 2 lineup itself looks like a set of specialized tools: Nova 2 Lite for fast everyday tasks, Nova 2 Pro (in preview) for more complex code and reasoning, Nova 2 Sonic for voice-based scenarios, and Nova 2 Omni (also in preview) as a multimodal model working with text, images, video, and audio. Nova Forge, however, is a different kind of statement altogether. AWS openly talks about an “open training” approach, where customers can start working from early checkpoints and mix their own datasets with curated Nova data — effectively creating a model variant tailored to their needs and then hosting it within AWS infrastructure.
Translated into business terms, what we are seeing is the normalization of a key idea: it is no longer just the knowledge base or search index that becomes “proprietary,” but the model itself — especially in domains where the cost of error, data leakage, or compliance failure is too high to rely on a universal API. 2026 increasingly looks like the year when competition will no longer be only between models, but between platforms that give companies the right to grow their own version of intelligence.
