Build custom AI apps with NVIDIA's Nemotron 3.5 and Kumo models on local infrastructure

NVIDIA has released a new set of open artificial intelligence models aimed at helping developers, researchers, and enterprises build applications tailored to local languages and industries. The release includes updates to language, robotics, and data processing tools, allowing organizations to customize software with local data. This expansion gives users a foundation they can deploy on infrastructure they control, supporting broader access to customizable AI capabilities.

In plain English

Open models provide the underlying code and weights that companies can modify to fit their specific needs. Unlike closed systems where users only access an application through a website or API, open models allow organizations to download the software and run it on their own servers. This approach gives enterprises greater control over data privacy and security, as they can keep sensitive information within their own networks rather than sending it to third-party providers.

The background

The new portfolio includes the NVIDIA Nemotron 3.5 Super VL and Nemotron 3.5 Bolt models, which support text and image processing. These models are built with contributions from the NVIDIA Nemotron Coalition, including European partners Black Forest Labs, H-Company, and Mistral. As earlier reporting noted, Microsoft leadership has framed AI models as insider risks, highlighting the tension between open innovation and security concerns. NVIDIA’s release offers tools that allow customization without sacrificing performance.

What changes now

Enterprises can now use the new NVIDIA Kumo family of models to handle structured business data more efficiently. Kumo helps applications forecast outcomes, score fraud risk, and recommend actions using relational and tabular data. Companies like Cognite, Coinbase, and Revolut are already using these models to accelerate development. Additionally, new Cosmos tools provide simulation checkpoints for robotics and autonomous vehicles, allowing developers to test behavior in virtual environments before deploying physical systems.

Key terms explained

Sovereign AI refers to the ability of nations or organizations to develop and deploy artificial intelligence systems using their own data and infrastructure. This concept emphasizes control over the entire lifecycle of AI models, from training to deployment. Teams usually assess whether their current architecture allows for local data processing and model customization. Checking vendor documentation for open-weight availability helps determine if a system supports sovereign requirements. Organizations often prioritize solutions that reduce dependency on external cloud providers for sensitive workloads.

What to do and how to stay safe: NVIDIA

• Review current AI vendor contracts to determine if data leaves your organization’s infrastructure during model training or inference.
• Evaluate whether your existing hardware can support the computational requirements of running large language models locally.
• Assess the security implications of using open-weight models, including the need for robust access controls and monitoring.
• Monitor vendor announcements for updates on data residency options and compliance certifications relevant to your industry.

General security guidance from the Vector Update newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Which companies are using the new Nemotron models?

Companies including Dassault Systèmes, Salesforce, SAP, Shopify, and WPP are working with Nemotron models to create custom applications.

What is the purpose of the NVIDIA Kumo models?

Kumo models help agents and applications forecast outcomes, score fraud risk, and recommend actions using structured business data.

How do the new Cosmos tools help developers?

Cosmos tools provide simulation checkpoints and optimization features that allow developers to test robotics and autonomous vehicle behavior before real-world deployment.