
Machenta, a London-based robotics startup founded by Imperial College London professors, raises $40 million in seed funding to train robots using simulated environments rather than real-world data, with Creandum and Union Square Ventures co-leading the round.
Machenta raises $40 million in seed funding to train robots without any real-world data. Creandum and Union Square Ventures co-led the round. That is a huge seed cheque for a London startup with just 11 people.
And it is a bold bet against the way the rest of the industry is building robots. Meet the Imperial College Team Behind Machenta Machenta was founded in 2025 by Edward Johns, an Imperial College London professor, and his former PhD student Vitalis Vosylius. Johns serves as CEO.
The idea behind Machenta took root about six years ago. Johns started asking a simple question: could robots pick up physical tasks as quickly as people do? That question shaped years of lab work at Imperial, which eventually gave birth to Machenta.
Early backers included Hoxton Ventures, and UK filings point to a smaller round of about £4.3 million in 2025. Now the company has far bigger firepower, raising $40 million in seed funding. How Machenta Trains Robots Without Real-World Data Machenta's model learns entirely inside computer-generated worlds.
These simulations mimic how objects move, slide and bump into each other. Johns told Business Insider the model was trained without a single pixel of real-world data.
The payoff? A robot can learn a new task from just one demonstration, with no extra training. You show it once, and it gets on with the job.
In demos, robots have tidied a table and packed objects into boxes on an assembly line. It is not perfect yet. Precise, complex movements are still tricky.
Early lab versions also needed about 30 minutes of extra data collection per task. Machenta wants to cut that step out completely. Now for the David-and-Goliath part.
Machenta's rivals are enormous. Physical Intelligence is valued at roughly $11 billion. FieldAI is reportedly raising at a $10 billion valuation.
Skild AI tops $14 billion and already sells similar technology. It reportedly makes $100 million in annual recurring revenue from more than 60 customers. Many of these players lean heavily on real-world data.
Machenta thinks that is a trap. Johns argues that piling up more data will not crack the problem alone.
In fact, nobody really knows how much data general-purpose robots will need. If simulation can close that gap, the economics of robotics could shift fast. Why This Fundraising Round Matters Collecting real-world robot data is slow and expensive.
You need hardware, people and endless hours of recordings. Simulation scales with compute instead, and compute keeps getting cheaper. That could let smaller teams train capable robots too.
That is why this round stands out. Machenta will use the cash to hire, buy more computing power and generate more simulated training data.
The goal is to tackle tougher, more precise tasks. Do not expect a product tomorrow, though.
The team is focused on research, with no near-term plans to commercialise. Early access to its model is not open yet, but robot builders can already register interest. For factories and warehouses, the promise is huge.
Imagine retraining a robot arm in seconds, not weeks. For investors, it is a sign that Europe can still compete in physical AI. Will simulation beat brute-force data collection