NVIDIA and Google Cloud Are Building Agentic AI Factories That Could Change Everything
NVIDIA and Google Cloud are pushing their decade-long partnership into bold new territory. At Google Cloud Next in Las Vegas this week, the two companies announced a sweeping set of upgrades designed to power the next generation of agentic AI factories systems capable of moving AI out of the lab and into real-world production environments, including factory floors and complex enterprise workflows.
The centrepiece of the announcements is the new A5X bare-metal instance, powered by NVIDIA Vera Rubin NVL72 rack-scale systems. According to NVIDIA, this architecture delivers up to 10x lower inference cost per token and 10x higher token throughput per megawatt compared to the previous generation, a significant leap for companies running large-scale AI workloads. These A5X instances can scale to up to 80,000 NVIDIA Rubin GPUs within a single site cluster, and up to 960,000 GPUs across multiple sites, making them among the most formidable AI infrastructure setups ever made commercially available.
For enterprises operating in regulated or data-sensitive industries, the partnership introduces confidential computing support. Google Gemini models can now run on NVIDIA Blackwell and Blackwell Ultra GPUs via Google Distributed Cloud a preview that allows organisations to keep sensitive prompts and fine-tuning data encrypted and shielded, even from infrastructure operators. This is reportedly the first confidential computing offering of NVIDIA Blackwell GPUs in any cloud environment.
On the agentic AI side, NVIDIA Nemotron 3 Super is now available on Gemini Enterprise Agent Platform, giving developers a cleaner path to deploying reasoning and multimodal models for agentic workflows. Google Cloud and NVIDIA are also rolling out a new managed reinforcement learning API built with NVIDIA NeMo RL, automating the heavy lifting of cluster sizing, failure recovery, and job execution so teams can concentrate on agent behaviour and model quality rather than infrastructure headaches. This infrastructure directly enables the rise of agentic AI factories, environments where AI agents manage complex, multi-step workflows autonomously.
The physical AI front is equally ambitious. NVIDIA Omniverse libraries and Isaac Sim, NVIDIA’s open-source robotics simulation framework, are now available on Google Cloud Marketplace, enabling developers to build accurate digital twins and validate robots in simulation before real-world deployment. NVIDIA Cosmos Reason 2 NIM microservices can also be deployed to Google Vertex AI and Google Kubernetes Engine, empowering vision AI agents to perceive, reason, and respond to physical environments, a foundation for the next wave of agentic AI factories in manufacturing and logistics.
The results are already showing. Schrödinger is compressing weeks-long drug discovery simulations into hours using NVIDIA accelerated computing on Google Cloud. Snap is cutting costs on large-scale A/B testing by migrating data pipelines to GPU-accelerated Spark on the same platform. OpenAI runs large-scale inference for ChatGPT on NVIDIA GB300 and GB200 NVL72 systems through Google Cloud. Meanwhile, CrowdStrike is using NVIDIA NeMo libraries to generate synthetic cybersecurity data and fine-tune models for faster threat detection and response.
More than 90,000 developers have joined the joint NVIDIA and Google Cloud developer community in just over a year, a signal of how rapidly this ecosystem is scaling. NVIDIA was also recognised at Google Cloud Next as Google Cloud Partner of the Year in both the AI Global Technology Partner and Infra Modernization Compute categories.
The agentic AI factories vision is no longer theoretical. With infrastructure that spans confidential computing, sovereign cloud deployment, physical AI simulation, and reinforcement learning at scale, NVIDIA and Google Cloud are laying the groundwork for AI systems that don’t just assist humans they run entire workflows, build better robots, and optimise factories in real time.
