Google Cloud and NVIDIA power microagi's embodied AI ambitions

Robotics deployment company microagi is folding Google Cloud and NVIDIA into its build-out, aiming to speed up how it trains AI models that let robots read and react to physical spaces. The Munich-based company, founded in 2025 with a research hub in Zurich and outposts in London and New York, will run its training and inference workloads on NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs and GB300 NVL72 rack-scale systems inside Google Cloud. It's not microagi's first contact with Google Cloud, CEO Bercan Kilic says, but this arrangement goes deeper, and Google's engineers have already helped the company roughly double its computational efficiency while cutting energy use per unit of work, with more tuning still underway.

The timing is notable. Just a week before this announcement, microagi closed a $55 million seed round, which it says is the largest seed round in German history, led by Hummingbird with Northzone, LocalGlobe, Village Global and redalpine joining in. What's stranger than the size is the speed: two days to close the pre-seed, three more for the seed, and a signed term sheet within three days of Hummingbird's visit. Kilic frames the quick close as investors simply catching up to work they'd been quietly watching for months, as the team showed off progress on data collection, compute, and robot training to labs and partners before any round was formally open.

At the center of the pitch is Atlas, microagi's platform for fine-tuning AI models on each customer's own operational data, producing robotics systems built for specific industrial jobs. It sits between a customer's infrastructure and frontier AI models, staying hardware- and model-agnostic so nobody gets locked into one vendor. Kilic, a former Formula 1 engineer at Red Bull Racing, says the pull toward robotics came from watching open-source Vision-Language-Action models emerge and realizing almost nobody in Europe was building for that frontier. He didn't set out to start a company. But the deeper he looked, the more obvious it became that Europe was short on three things: large-scale robotics data, serious compute capacity, and the infrastructure to actually train and deploy embodied AI.

That gap is what drives microagi's insistence on keeping workloads on European infrastructure whenever it can. Customers are European manufacturers guarding sensitive production data, and running under GDPR gives them extra assurance even when they trust the vendor. Kilic also frames it as a longer bet: if advanced chips ever face export restrictions or geopolitical friction tightens, Europe will only have the compute it already built at home. He argues the next 18 months matter enormously for Europe to invest in energy, data centers, and compute — not just for microagi, but for the wider ecosystem around it. Left unaddressed, he warns, the technology gap between Europe and the US and China could end up wider than the gap between Europe and developing economies today.

NVIDIA's Tobias Halloran, Director of EMEAI Startups, describes robotics as one of AI's most demanding frontiers, needing massive physical-world datasets and full-stack infrastructure to turn models into working machines. microagi's answer is to control more of that stack itself: collecting its own robotics data, training foundation models at scale, then fine-tuning per customer while letting that customer keep ownership of both data and resulting models. Kilic contrasts this with setups where customer data quietly flows back into shared foundation models. For now, the company says its job is straightforward execution — it has customers, data, and compute, and the next phase is just scaling deployments while pushing for more European compute capacity underneath it all.