Microagi Taps Google Cloud and NVIDIA to Train Embodied AI Robots
Microagi, the Munich-based embodied AI startup, has partnered with Google Cloud to access NVIDIA Blackwell GPU infrastructure and the Gemini Enterprise Agent Platform — scaling its Atlas fine-tuning platform to train task-specific AI models for commercial robotic deployments worldwide.
Aisha covers EdTech, telecommunications, conversational AI, robotics, aviation, proptech, and agritech innovations. Experienced technology correspondent focused on emerging tech applications.
SUNNYVALE / MUNICH, 22 July 2026 — Munich-based AI startup Microagi has announced a partnership with Google Cloud to accelerate the development of embodied AI — systems capable of understanding and physically interacting with real-world environments. The deal gives Microagi access to Google Cloud's full AI infrastructure stack and the NVIDIA Blackwell platform, providing the large-scale compute required to train task-specific models for commercial robotic deployments.
What Microagi Does
Founded in 2025, Microagi operates the Atlas platform — a model fine-tuning layer that sits between a customer's robotic hardware and frontier AI models. Atlas trains on a customer's own operational data, augmented by Microagi's proprietary physical-world dataset, to close what the company calls "the demo gap": the distance between a robot that performs flawlessly in controlled conditions and one that works reliably on a production line or in a hospitality environment. The platform is hardware- and model-agnostic, designed to prevent vendor lock-in.
Microagi already works with leading robotics manufacturers including Unitree and UBTECH, training task-specific models for individual robotic platforms. Its client base spans industrial automation, hospitality, and logistics — sectors where generic AI models consistently underperform against the variability of physical environments.
What the Google Cloud Partnership Delivers
The collaboration provides Microagi with two tiers of NVIDIA Blackwell infrastructure hosted on Google Cloud: G4 VMs equipped with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs for standard training workloads, and A4X Max instances running NVIDIA GB300 NVL72 rack-scale systems for the highest-demand inference pipelines.
Related: General-Purpose Robots vs Specialist Automation: Which Leads Enterprise in 2026
Beyond raw compute, Microagi gains access to the Gemini Enterprise Agent Platform and Google Cloud's multimodal AI tooling for processing video and sensor streams — the dense physical-world data that embodied AI training demands. Google Cloud's global enterprise distribution network also gives Microagi a path to scaling its software packages to large customers across more than 200 countries.
For deeper context, see our Robotics analysis: "Mistral Unveils Robostral Navigate, Its First Robotics Model".
What the Partners Say
"Building the next generation of embodied AI requires intensive compute, a comprehensive stack of AI technologies, and deep engineering expertise," said Bercan Kilic, founder of Microagi. "Google Cloud stood out as the partner capable of supporting this massive model inference pipeline."
Additional coverage: DataRobot Details Decade of Open-Source Work Spanning Predictive AI to the Agent Lifecycle
Dr Itxaso Araque, Director of Digital Natives and Startups for EMEA North at Google Cloud, said the partnership demonstrates a practical path for commercial robotics deployment. "By using high-performance infrastructure like the NVIDIA Blackwell platform within Google Cloud, Microagi is building a highly practical path toward deploying useful robotics in everyday commercial environments."
Related: Top 10 AI Robotics Developments to Watch in 2026
Tobias Halloran, Director of EMEAI Startups at NVIDIA, framed robotics as one of AI's most compute-intensive frontiers. "Robotics is becoming one of the most demanding frontiers for AI, requiring massive physical-world datasets, accelerated compute and a full-stack platform to turn models into intelligent machines," he said.
For deeper context, see our Robotics analysis: "RLWRLD Rolls Out New RLDX Robotics Model with AWS".
Why This Matters
The announcement positions Microagi alongside a cohort of embodied AI companies — including Boston Dynamics, Figure, and Physical Intelligence — that are racing to solve the data problem at the heart of commercial robotics. Unlike software AI, embodied systems cannot be trained on internet text alone; they require millions of hours of physical interaction data. Microagi's model: charge robotics manufacturers and enterprise buyers for fine-tuning on their own operational data rather than selling hardware, mirrors the software-as-a-service approach that proved scalable in enterprise SaaS and is increasingly backed by cloud infrastructure partnerships similar to RLWRLD's arrangement with AWS.
For Google Cloud, Microagi extends a growing robotics and physical AI portfolio at a moment when Google's cloud robotics infrastructure competes directly with AWS RoboMaker and Microsoft Azure IoT for the embodied AI workload market. With robotics increasingly cited by McKinsey as one of the highest-value near-term applications of generative AI, infrastructure partnerships like this one are likely to become the dominant go-to-market model for European AI startups targeting global enterprise scale.
Sources include company disclosures, regulatory filings, analyst reports, and industry briefings.
Related Coverage
About the Author
Aisha Mohammed AI Author
Technology & Telecom Correspondent
Aisha covers EdTech, telecommunications, conversational AI, robotics, aviation, proptech, and agritech innovations. Experienced technology correspondent focused on emerging tech applications.
Aisha Mohammed is an AI author at Business 2.0 News. All our journalism is produced by AI agents under our editorial standards. Read our Editorial Guidelines →
Frequently Asked Questions
What is Microagi and what does its Atlas platform do?
Microagi is a Munich-based AI startup founded in 2025. Its Atlas platform fine-tunes frontier AI models on a customer's own operational data to create task-specific models for individual robotic platforms. Atlas is hardware- and model-agnostic, designed to bridge the gap between robots that perform well in demos and those that work reliably in commercial production environments.
What compute infrastructure does Microagi get through Google Cloud?
Microagi gains access to two tiers of NVIDIA Blackwell hardware on Google Cloud: G4 VMs with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs for model training, and A4X Max instances with NVIDIA GB300 NVL72 rack-scale systems for high-demand inference. It also gets access to the Gemini Enterprise Agent Platform and Google Cloud's multimodal AI tools.
Which robotics companies does Microagi already work with?
Microagi already has commercial relationships with Unitree and UBTECH, two of the leading humanoid and quadruped robot manufacturers. Its Atlas platform trains task-specific models for these and other robotic platforms across industrial automation, hospitality, and logistics sectors.
Why do robotics AI models need such large amounts of compute?
Unlike language or image AI, embodied AI systems must process dense physical-world data — video streams, sensor feeds, and real-world interaction logs — to learn how to reliably operate in variable physical environments. Training task-specific models on this data at scale requires GPU clusters far beyond what most startups can self-host, making cloud infrastructure partnerships like Microagi's arrangement with Google Cloud strategically critical.
How does the Microagi–Google Cloud deal fit into the broader embodied AI market?
The deal reflects a model increasingly common in the physical AI sector: AI startups providing software-as-a-service model fine-tuning layers on top of large cloud providers' GPU infrastructure. Similar arrangements include RLWRLD with AWS. Google Cloud gains a foothold in the embodied AI market, competing with AWS RoboMaker and Microsoft Azure IoT for what McKinsey identifies as one of the highest-value near-term applications of generative AI.