NVIDIA Open Sources Aerial Software to Accelerate AI-Native 6G

NVIDIA is delivering the telecom industry a major boost in open-source software for building AI-native 5G and 6G networks.

NVIDIA Aerial software will soon be released as open source, making it available on a variety of NVIDIA platforms, including on NVIDIA DGX Spark. With open-source software and a powerful and accessible supercomputer to run it all, AI-RAN and wireless researchers can go from rapid prototyping to product development in hours versus months or years.

That means next-generation AI-native mobile networks can be built at the pace of AI for open, widespread collaboration, unlike previous generations where innovation was constrained by access to licensed software and proprietary hardware.

Rewriting the Wireless Playbook With Open Source Open source is reshaping industries by breaking down barriers to innovation and accelerating collaboration, enabling breakthroughs that would be impossible under closed, proprietary systems. NVIDIA is a leading contributor to open source, including for the telecom industry. Previously open-sourced NVIDIA Sionna software has already surpassed 200,000 downloads and 500 citations.

Now, NVIDIA is open sourcing its Aerial software, including Aerial CUDA-Accelerated RAN, Aerial Omniverse Digital Twin (AODT) and the new Aerial Framework.

These resources - previously limited to a small group - are expected to be available on GitHub under Apache 2.0 licensing starting this December, with AODT release in March 2026, empowering developers to build full-stack, AI-native 5G and 6G RAN solutions. They can experiment and build AI-native network solutions without restrictions, accelerating the transition from research to real-world deployment.

The upcoming Aerial open-source release is packed with capabilities, including:

Aerial Framework for converting Python code into high-performance CUDA code to run on NVIDIA Aerial RAN computer platforms.

AI-powered neural models, such as advanced channel estimation for significantly improving wireless performance.

A dApp framework to provide third-party applications with access to real-time physical layer data through secure application programming interfaces. Developers can deploy AI-powered dApp algorithms that can modify RAN behavior in real time.

Customizable pipelines with modules that developers can selectively modify or replace with their own code to build full-stack RAN software.

Such features have enabled the first made-in-America AI-native wireless stack, showcasing early 6G applications including spectrum agility and integrated sensing and communications.

This move expands access to powerful CUDA-accelerated wireless software and boosts the global movement toward AI-native 5G and 6G solutions. It opens doors to developers beyond the telecom industry, making it easy for them to build new applications for mobile networks, including agentic and physical AI applications that need mission-critical performance.

This shift amplifies U.S. leadership in open source and fundamentally rewrites the way the telecom industry innovates - sparking knowledge-sharing and an implementation-first approach at a global scale.

DGX Spark: The Desktop Supercomputer for Wireless R&D NVIDIA DGX Spark, the world's smallest AI supercomputer, is now available for AI-native 5G and 6G research. It delivers the performance to run NVIDIA Aerial or Sionna software in a cost-effective small footprint. Whether procured through manufacturers or accessed via NVIDIA DGX Cloud, DGX Spark empowers teams to prototype complete wireless networks and continuously train and refine their AI models using real-world data gathered across diverse radio environments.

The NVIDIA Sionna Research Kit now supports DGX Spark, in addition to NVIDIA Jetson AGX Orin, offering an AI-native 6G lab in a box for rapid prototyping. It brings together every component, from user equipment and antennas to radio systems and core networks, that developers need to research and validate AI/ML algorithms' performance over the air - and to do so in one afternoon, from unboxing to setting a live 5G network. Researchers can test their algorithms anywhere with the ultraportable kit, instead of being confined to a lab.

The NVIDIA Aerial Testbed now supports DGX Spark in addition to NVIDIA GH200 Grace Hopper Superchips, enabling over-the-air testing of CUDA-accelerated, full-stack 5G and 6G networks. Designed to accelerate commercial product development, it provides seamless integration between network digital twin and live deployment.

Dell Technologies is introducing Dell Pro Max with GB10, a DGX Spark-based system designed to empower researchers and developers across a wide range of AI applications, including the demanding needs of global telecom research. The platform delivers proven performance for handling intensive 5G and 6G workloads, offering a stable and powerful environment for testing, simulation and validation while enabling researchers to tackle the complex challenges of 6G.

Collaborators Building AI-Native 6G on NVIDIA The NVIDIA AI Aerial portfolio - spanning accelerated computing platforms, software libraries and tools - gives developers the means to build, train, simulate and deploy full-stack AI-native RAN systems faster than ever.

Thousands of wireless innovators around the world are already tapping into the NVIDIA AI Aerial portfolio for research and development, including leading U.S. institutions Northeastern University, Virginia Tech, Arizona State University and DeepSig, and researchers from WINSLab and LIDS at MIT, who are working on AI-driven breakthroughs that will define 6G technology and shape global standards.

The AI-RAN Alliance is shaping the core architecture of AI-native wireless networks, bringing together over 100 telecom industry pioneers through work groups, labs, benchmarking and live demonstrations - many of which have bee

LINK:	https://blogs.nvidia.com/blog/open-source-aerial-ai-native-6g/...
	See more stories from nvidia

More from Nvidia