AWS launches Nova Forge, a $100,000/year service allowing clients to customize Amazon's AI models at various stages of training and refine open-weight models
Jordan Novet / CNBC :
Context & Ripple Effects
Nova Forge extends AWS’s Nova lineup from serving multimodal models in Bedrock—following Nova Premier’s Bedrock availability—to offering customers more control over how models are adapted. It also builds on AWS’s longer-standing interest in portable, open ML tooling, reflected in the earlier Neo-AI open-source project.
The launch places model customization alongside AWS’s core infrastructure role, shortly after OpenAI committed major spending on AWS compute. That combination makes the model layer a more direct part of AWS’s cloud sales proposition.
First-order effects
- Customers able to meet the $100,000 annual price can customize Amazon models at multiple training stages or refine open-weight models through an AWS-managed service.
- AWS gains a higher-value offering around Nova, shifting its pitch from model access alone toward tailored model development for enterprise users.
Second-order effects
- Cloud AI rivals face added pressure to package model adaptation, hosting, and training support as integrated services rather than leaving customers to assemble those pieces separately.
- The service gives buyers a formal procurement path for customized and open-weight-model work, potentially concentrating related compute and deployment spending within AWS.
Third-order effects
- If adoption broadens, cloud platforms may compete increasingly on control over the model-development workflow—customization, training infrastructure, and deployment—rather than on foundation-model access alone.
- A paid, managed customization layer could segment the market between enterprises that can fund bespoke model work and users relying on standard hosted models.
The trend: AI cloud providers are turning foundation models into configurable enterprise platforms that bundle model adaptation with the infrastructure on which those models are built and run.