MUNICH, July 22, 2026. One week after announcing the largest seed round in German history, microagi is formalising a compute partnership with Google Cloud and NVIDIA. Atlas, our data and deployment platform, now trains and serves its models on NVIDIA Blackwell infrastructure on Google Cloud: GB300 NVL72 rack-scale systems (A4X Max instances) and RTX PRO 6000 Blackwell GPUs (G4 VMs), alongside the Gemini Enterprise Agent Platform and leading AI models for processing video and other multimodal data. Engineering teams from all three companies work together directly.
Why compute decides
Language models had the internet to learn from. There is no internet for robotics. The data robots learn from has to be captured in the physical world, inside live production, and turned into models through training. Compute is the ceiling on how fast that can happen.
The demand is also unusually spiky. Every deployment follows the same curve: months of data collection at a customer’s site, then a point where the dataset is ready and a model must be trained for that specific plant. At that moment, the compute a single customer needs jumps from modest to frontier scale. Multiply that by a growing customer base and the infrastructure question stops being a detail of operations and becomes the business.
“Robots learn from data, and data is useless without the compute to train on it,” said Bercan Kilic, founder and CEO of microagi. “This partnership gives us scale, and it gives us engineering depth we could not build alone at this stage: NVIDIA optimises down to the card, Google Cloud across the whole machine. Our customers see the difference in what a deployment costs.”
What the partnership includes
Beyond raw capacity, the three engineering teams work on the same problem from different ends. A single training or annotation job runs through thousands of steps, each costing seconds of machine time, and those seconds add up to what an industrial customer ultimately pays. NVIDIA engineers optimise performance at the level of the GPUs; Google Cloud engineers optimise the surrounding stack, from VM configuration to CPU utilisation to job orchestration. Since the collaboration began, microagi has roughly doubled the work it gets from every unit of energy.
That efficiency matters beyond our own margins. Many European industrial companies are under real economic pressure, and the cost of training and running their models flows directly into the price of deploying robots. Cheaper compute per unit of work means automation that pays for itself sooner.
Most of microagi’s training runs on Google Cloud capacity in the Netherlands, and the company requests European capacity for its European customers as a matter of policy.
“Physical AI is one of the hardest problems in the field, and after seeing what microagi is doing, it was clear we would put our best resources behind it,” said Dr. Marianne Janik, VP EMEA North, Google Cloud. “We support microagi with compute, with the Gemini Enterprise Agent Platform and with leading models for processing multimodal data. Just as important, we deliberately reserve compute for startups in Europe, because access to compute decides whether Europe’s data can be processed in Europe. Physical AI is an area where Europe, and Germany in particular, has developed real ambition.”
“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,” said Tobias Halloran, Director of EMEAI Startups at NVIDIA. “By running on NVIDIA Blackwell powered instances on Google Cloud, microagi can scale the training and deployment of embodied AI systems for commercial and industrial environments.”
The European part
Compute is becoming as strategic as energy. It increasingly determines which economies can automate, which factories keep running, and who builds the machines that will do the world’s work. Europe currently owns very little of it: neither the chips nor the hyperscale capacity its own companies run on.
microagi’s position is that demand is the fastest way to change that. China installed 295,000 factory robots last year; all of Europe installed 85,000. Europe still holds two centuries of industrial knowledge, and turning it into working machines takes compute at frontier scale, on terms Europe can rely on.
“The market is usually fair: where there is demand, there is supply,” said Kilic. “All we are trying to do is create more demand for European compute, because the industrial base we are trying to rescue is European. If Europe does not start buying and building serious compute in the next eighteen months, the gap stops closing.”
Working with us
Industrial companies interested in Atlas can get in touch through the Atlas contact form. Open roles are listed at microagi.ai/careers.
About microagi
microagi, founded in 2025, is a robotics deployment company. Its platform, Atlas, fine-tunes models on a customer’s own operations, drawing on microagi’s proprietary data corpus, and closes the gap between the robot that impresses in a demo and the robot that does reliable work on the line. Atlas is hardware- and model-agnostic: a neutral layer between a customer’s infrastructure and frontier AI models, designed to prevent lock-in to any single vendor. microagi is headquartered in Munich, with a research hub in Zurich and offices in London and New York.
About Google Cloud
Google Cloud offers a powerful, optimized AI stack that enables organizations to transform their business for the Agentic Era. It includes AI infrastructure, leading models like Gemini, data management capabilities, multicloud security solutions, developer tools and platform, as well as agents and applications. Customers in more than 200 countries and territories turn to Google Cloud as their trusted technology partner.
https://www.microagi.ai/articles/google-cloud-nvidia
