The archive · AI & Models · Strategic decision · 2025–2026
Microagi bets real factory data, not new hardware, will make robots work; $55M seed
Munich's Microagi raised Germany's largest-ever seed ($55M) to sell Atlas, a model-agnostic layer that fine-tunes existing robots on real factory data.
Microagi
What the business is
Microagi sells Atlas, a hardware- and model-agnostic platform that records operations in customer factories and fine-tunes existing robot models to plant-specific tasks; its consumer arm Shift gathers household footage for home-robot training.
How it started
Bercan Kilic, an aerodynamics engineer who joined Red Bull Racing in 2023, left Formula One with co-founders from Red Bull and Mercedes-AMG Petronas to found Microagi in Munich around September 2025. Their premise: robots impress in demos but fail in production because training data from real factories and homes is scarce, low-quality and proprietary.
What happened
The company built Atlas around dedicated recording hardware and a secure ingestion pipeline that curates factory-floor data and fine-tunes frontier robot models, then keeps refining them with a reinforcement loop on site. By July 2026 five unnamed customers across automotive, logistics and food were feeding data through Atlas, with one preparing to deploy robots in a factory, and Microagi had opened a Zurich research hub plus London and New York offices. On 2026-07-16 it announced a $55M seed led by Hummingbird with Northzone, LocalGlobe, Village Global and Redalpine — about ten months after founding — which Sifted and automate.org reported as Germany's largest seed round ever. Its consumer arm Shift also drew viral attention in June 2026 by offering New Yorkers free apartment cleanings filmed with body-worn cameras, plus private-chef recordings in San Francisco, to collect first-person footage for training home robots.
No ending yet — it is still running.
Background
Microagi was founded in Munich around September 2025 by Bercan Kilic — a Red Bull Racing aerodynamics engineer who joined the team in 2023 — with co-founders drawn from Red Bull and Mercedes-AMG Petronas Formula One teams. Their diagnosis was that industrial robotics has a data problem: demo robots impress, but production robots fail because most training data is not real factory-floor footage, and robot labs cannot buy what does not exist.
Its answer, Atlas, is deliberately not another robot or foundation model. Atlas is a hardware- and model-agnostic data and deployment layer: Microagi engineers place dedicated recording hardware in a customer's plant, ingest and curate edge-case data, and fine-tune whichever robot model the customer already uses, then keep refining it in a reinforcement loop so performance improves month over month. By July 2026 five unnamed customers spanning automotive, logistics and food were collecting data through Atlas, with one preparing to deploy robots in a factory.
On 2026-07-16, about ten months after founding, Microagi announced a $55M seed round led by Hummingbird with Northzone, LocalGlobe, Village Global and Redalpine — reported by Sifted and automate.org as the largest seed round in German startup history. The company also expanded to a Zurich research hub and London and New York offices, and its consumer arm Shift made news in June 2026 by offering New Yorkers free cleanings filmed with body-worn cameras and private-chef cooking sessions in San Francisco to capture first-person household footage for training home robots.
The bet is unproven as of September 2026: Atlas has named no customers publicly, home-robot data collection still faces questions about consent and privacy, and the startup must show that fine-tuning on operational data reliably closes the gap between robot demos and dependable production work. But the record seed round signals investors are paying for the data wedge itself.
What has to be true
- Data scarcity: LLMs trained on the web, but robot models lack filmed examples of people actually doing tasks, so Microagi made collecting and curating that footage its core asset.
- No hardware bet: staying hardware- and model-agnostic let Microagi avoid the capital sink of building robots and made every robot vendor a potential partner instead of a competitor.
- Reinforcement-loop economics: each extra month on a customer's line makes the fine-tuned model better, creating switching costs and recurring value that grows with time.
- Consumer data channel: Shift's free-cleaning and private-chef offers generated viral attention and first-person household data that robot labs cannot license elsewhere.
What can be applied
Microagi bet curated real-world data is robotics' scarce resource, so it monetized a neutral fine-tuning layer instead of building robots — and raised Germany's largest seed on that wedge.
Aftermath
As of 2026-09-02 Microagi is a live, privately held startup about a year old, freshly funded with Germany's largest seed ($55M) and expanding from Munich (HQ) with research in Zurich and offices in London and New York. It reports five industrial customers collecting data through Atlas across automotive, logistics and food, one preparing factory deployment, and no named customers or disclosed revenue. Shift's June 2026 free-cleaning and chef campaigns drew broad coverage and debate over whether data-for-services swaps are a fair deal, a question the company has not fully answered.
Sources
- Microagi nabs $55M to teach factory robots how to work
- Munich robotics startup Microagi lands $55 million seed round to expand Atlas deployment platform across major industrial groups
- Robotics raises: BRINC, Microagi, Monumental, Walden Robotics, Xynova
- Microagi获5500万美元融资，用AI训练工厂机器人
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card