← BACK TO FEED
NVIDIAGoogle CloudJAXinferencedeveloper tools

NVIDIA and Google Cloud Are Arming 100,000 Developers With AI Tools — Here's What's Actually New

NVIDIA and Google Cloud are expanding their joint developer community—now over 100,000 strong—by introducing new learning resources such as a JAX-on-NVIDIA-GPUs learning path, an NVIDIA Dynamo inference codelab, and monthly developer livestreams. Together, they are equipping developers with tools and frameworks to build and deploy production-ready AI applications, including multi-agent systems using Google DeepMind's Gemma models, NVIDIA Nemotron, and optimized inference on Google Kubernetes Engine. The partnership also prioritizes responsible AI development through collaboration on Google DeepMind's SynthID watermarking technology integrated with NVIDIA Cosmos world foundation models.

Google I/O is the natural habitat of developer announcements, and this year NVIDIA and Google Cloud are using the occasion to flag updates to their joint developer community, which now claims over 100,000 members. The community, which launched at last year's Google I/O, targets developers, data scientists and ML engineers who want hands-on experience with NVIDIA's stack running on Google Cloud infrastructure.

So what's actually new? Three things worth noting. First, a learning path focused on running JAX workloads on NVIDIA GPUs. Second, a codelab covering inference optimisations using NVIDIA Dynamo on Google Kubernetes Engine. Third, monthly developer livestreams, which is a low-effort addition but presumably useful for keeping the community from going stale.

The JAX work is arguably the most technically interesting piece here. NVIDIA and Google Cloud have been aligning around the framework to let developers scale JAX workloads from single-GPU experiments up to multi-rack deployments without the experience falling apart. Google Cloud's MaxText framework sits on top of these same JAX optimisations to train large models on NVIDIA GPUs via AI Hypercomputer. That's a reasonably coherent stack, even if it takes some squinting to see where one company's contribution ends and the other's begins.

On the model side, developers are being pointed toward combinations of Google DeepMind's Gemma 4 models, NVIDIA's Nemotron open models, and Google's Agent Development Kit. The suggested deployment target is Google Cloud G4 VMs running NVIDIA RTX PRO 6000 Blackwell GPUs, either via Cloud Run or spot instances. For data work, NVIDIA's cuDF library gets a mention alongside Google Colab Enterprise and Dataproc.

There's also a responsible AI angle worth a paragraph. NVIDIA was apparently the first industry partner to integrate Google DeepMind's SynthID watermarking technology, which embeds digital watermarks into AI-generated content. The pairing here is with NVIDIA's Cosmos world foundation models, which handle 3D simulation for robotics and autonomous systems. The idea is that if your robot is generating imagery or video, SynthID can stamp it so you at least know where it came from. Whether that meaningfully addresses trust concerns in physical AI deployment is a longer conversation, but it's not nothing.

The broader partnership context: at Google Cloud Next, the two companies announced expanded collaboration including NVIDIA's upcoming Vera Rubin-powered A5X instances and work with Gemini models. Customers named as using the combined stack include OpenAI, Salesforce, Snap and CrowdStrike, which is a decent roster even if it reads a bit like a conference badge wall.

For developers already working in this ecosystem, the new codelabs and learning paths drop next month. For everyone else, the headline is that NVIDIA and Google Cloud are doubling down on making their joint infrastructure story accessible to people who actually build things, rather than just the enterprises buying it in bulk.