Physical AI Unlocking scalable heavy industry
In short
No contractor would trial its robot dozer, so the start-up bought one
Gemma Bloemen and Camilla Dolan asked founders and investors whether physical AI can revive Europe's heavy industry, and the answers kept returning to customers who will not change.
A start-up developing an electric autonomous dozer could not find an excavation contractor willing to trial it. So it bought one, a business with 100 million in revenue, and then a steel fabrication plant to control its bill of materials. A later-stage investor told the story to show how far physical AI companies now go to reach a customer. Demand for robots in Europe's heavy industry is real. Getting a customer to change how it works is the hard part.
Camilla Dolan of Eka Ventures and Gemma Bloemen of Systemiq Capital asked whether physical AI is the answer for heavy industry in Europe, or whether everyone is running a little ahead. The demand case came first. A construction-robotics founder said Europe must fill 4.2 million construction jobs by 2035, and 4.1 million of them replace people leaving the industry. European customers have felt that labour crunch first, which makes them early adopters. The founder also saw an edge in talent. At-will employment in the US is a real problem for companies whose people need years to build up deployment know-how.
Spreadsheets and a wrong formula
The stories about customers were much alike. A founder selling wastewater analytics had industrial clients who offered to send data by the hour and did not know what an API was. A deep-tech investor described a portfolio company that can cut up to 10–15% of the food waste on a line carrying 5,600 kilos of tomatoes a day. The customer still asked whether it really needed the system now. The construction founder had watched one of the larger US home builders puzzle over a spreadsheet for a house it had built many times. The formula was wrong. The start-up recalculated it at the press of a button.
A participant with a decade in metal manufacturing said money is not what holds job shops back. Profitable shops hold plenty of cash and would happily spend half a million or a million to cut the cost per part. What they care about is technical utilisation, which averages 30–50% across the industry. Even a shop with 50 million in revenue a year runs on guesswork, with people calculating by hand as they did 40 years ago. Until shops turn that guesswork into a system, the founder said, they cannot tell whether a cobot made things better or worse.
The budget is delayed by 12 months and that's 12 months runway.
Innovation by stealth
The construction founder's answer is to ask customers to change nothing. Home builders run on thin margins, often do not own the land they build on and do not want to own an asset. So the start-up takes a builder's existing designs, delivers the output the builder already knows, adds value on top, charges less than the builder pays today and asks for no capital. "I call it stealth innovation to the builders that I work with," the founder said. What to integrate comes later, once both sides have learned what is worth owning.
A corporate venture investor from a large industrial reached the same point with machines. Industrials have capital tied up in fleets of excavators they already own. A start-up that asks them to buy a new autonomous excavator has a weaker offer than one that retrofits the old fleet and keeps the upfront spend small. Asked whether a retrofit kit counts as physical AI, the investor said yes, since the AI is the perception and the software.
Who pays for the wait
The later-stage investor said few physical AI companies meet late-stage criteria, and by the time they do, the rounds are too large for the ownership such funds want. The dozer company raises a harder question, of how to value a technology developer that now runs excavation services. One of the hosts said the models that seem to work are vertically integrated ones such as Octopus Energy, which began that way and now sells its software to other companies. A show of hands still suggested some reluctance among investors to back that model.
Two investors described how they judge it. One said it would be foolish to dismiss forward-deployed engineering, given how many successful companies have used it, but wanted proof that a start-up can move customers onto a platform that no longer depends on custom engineering, and that each rollout gets shorter. The other warned that a full-stack company pays for engineers and operations every month while customers deliberate, and that investors can easily pay too much for early signs of adoption. Expect at least one bridge round, they said, and that is the best case. They had lived it with one portfolio company. "The budget is delayed by 12 months and that's 12 months runway."
That investor still thought the company might make a lot of money in the end. The construction founder had chosen the other road after watching vertically integrated construction start-ups take on that much capital and operational risk and scale very slowly. Nobody offered a rule for who should fund the years between a working robot and a customer ready to use it. Europe, Gemma said at the close, is at the very beginning of this journey.
This Ripple was hosted by Gemma Bloemen (Systemiq Capital) and Camilla Dolan (Eka Ventures) at The Drop 2026 on 16 September.