NVIDIA and Google Cloud announced an expansion of the joint AI developer community they run together, timed to this year's Google I/O[1]. The community now has more than 100,000 members, and the new additions include a learning path for running JAX on NVIDIA GPUs, a NVIDIA Dynamo codelab focused on inference optimization, and monthly developer livestreams[1]. The aim is to give builders everything they need to develop AI agents, learning resources included.

A New Lineup for a Community of More Than 100,000 Developers

Launched at last year's Google I/O, the community brings together developers, data scientists, and machine learning engineers who want to sharpen their AI skills using the latest NVIDIA and Google Cloud technologies[1]. Drawing on the full-stack NVIDIA AI platform on Google Cloud, it has supported their work through learning paths (structured curricula), hands-on labs, and events[1].

New this year are a learning path for running the machine learning library JAX on NVIDIA GPUs, a codelab (hands-on practice) for NVIDIA Dynamo—software that streamlines inference—and monthly developer livestreams[1].

According to NVIDIA, over the past year the community has become a go-to hub for people building AI, yielding results such as production-ready RAG (retrieval-augmented generation, a method that references external data to compose answers) applications on Google Kubernetes Engine (GKE) and setups that provide observability into how agents are running[1]. Builders are also prototyping inference that combines on-premises and cloud environments for real-world uses such as sports analytics and enterprise data pipelines[1].

Building with Gemma 4, Nemotron, and Open Frameworks

The two companies offer learning materials and labs that combine NVIDIA's libraries, open models, and tools with Google Cloud's AI platform so developers can build production-quality AI applications quickly[1]. For example, developers can use NVIDIA's cuDF library, which accelerates data processing, in Google Colab Enterprise or Dataproc, or combine Google DeepMind's open model Gemma 4, NVIDIA's open model Nemotron, and Google's Agent Development Kit (ADK) to run multi-agent applications in which several AIs work together[1]. For the runtime, they use Google Cloud's G4 virtual machines powered by NVIDIA RTX PRO 6000 Blackwell GPUs, on Cloud Run or as spot instances[1].

For the open framework JAX, the companies are deepening their collaboration so that everything from single-GPU experiments to large configurations spanning multiple racks can run on NVIDIA's AI infrastructure with consistent performance and experience[1]. The MaxText framework, which trains large models efficiently, uses these JAX optimizations on Google Cloud AI Hypercomputer[1]. On the inference side, NVIDIA Dynamo running on GKE streamlines large-scale inference such as MoE (Mixture of Experts) models that selectively use multiple specialist models[1]. The learning path for running JAX on NVIDIA GPUs and the inference codelab covering Dynamo on GKE will both become available to community members next month[1].

Backing "Responsible AI" with SynthID and Cosmos

AI agents are increasingly built by combining multiple models that reason, plan, and act on behalf of users[1]. Amid this shift, NVIDIA notes that trust and transparency—being able to understand where generated content comes from and how it works—are foundational[1].

NVIDIA was the first industry partner to collaborate with Google DeepMind on SynthID, an AI watermarking technology[1]. SynthID embeds unobtrusive digital watermarks into AI-generated content and helps preserve the integrity of the outputs created by NVIDIA's Cosmos family of foundation models[1]. Cosmos is a set of world foundation models that provide 3D spatial perception and simulation capabilities for physical AI such as robots and autonomous machines, and it is available on build.nvidia.com[1]. Combining the two makes it easier to deploy agentic applications responsibly—from cloud to edge to real-world environments—while preserving the transparency of images and video[1].

A Growing Full-Stack Platform

This year's Google I/O puts the spotlight on new agentic experiences and developer tools[1]. NVIDIA and Google Cloud are focused on providing the infrastructure, software, and learning resources developers need to make the most of them, and they say the skills and tools gained in the community can scale directly from prototype to enterprise-grade production[1]. At the recent Google Cloud Next, the companies also expanded the full-stack platform that supports training, deploying, and operating agents[1]. This work includes A5X instances powered by NVIDIA's next-generation Vera Rubin design and Google DeepMind's Gemini models, and it is being used by leading AI labs and enterprises such as OpenAI, Thinking Machine Labs, Schrödinger, Salesforce, Snap, and Crowdstrike[1].

Summary

NVIDIA and Google Cloud have added JAX and NVIDIA Dynamo learning tracks to their joint developer community and further enriched the environment for building AI agents, including G4 instances powered by RTX PRO 6000 Blackwell alongside Gemma 4 and Nemotron. From SynthID watermarking and responsible-AI work using Cosmos to A5X instances based on Vera Rubin, their "full-stack" orientation—making everything from learning to production continuous—is clear. The new learning path and codelab are scheduled to become available next month.

Source: https://blogs.nvidia.com/blog/google-cloud-developer-community-ai-builders/