The first industrial robot went to work at a General Motors plant in New Jersey in 1961. Machines have moved things on factory floors ever since, and over the decades learned to sense and decide. The subject now has a new name, Physical AI, and a new scale of funding to go with it.
The engineering underneath is often excellent. The investment cases built on it are often weaker than they look, and the weakness tends to sit in the same place.
The demonstration
A demonstration is evidence of one thing. The machine works under conditions chosen by the people who built it, operated by people who understand it, for as long as the camera runs.
An automation business needs evidence of several more things. The machine has to install at a customer’s site, among equipment it was never designed around. The customer’s own operators have to run it. It has to keep running through shift changes, dust and a maintenance team nobody consulted. The company has to service it at a cost that leaves a margin. Then all of that has to happen again at the next site, with less of the founders’ time.
Customers buy automation expecting fifteen years of service, and they will use it in ways nobody tested. The value of the business lies in that span, a long way from the demonstration.
The ladder
The evidence comes in rungs, and each is harder to fake than the one before.
| Rung | What it shows | Where it usually appears |
|---|---|---|
| Works once | Controlled demonstration | Seed and Series A decks |
| Works there | Running at a customer site, run by the customer’s staff | Pilots, often subsidised |
| Works the nth time | Site after site, the same result, each quicker and cheaper than the last, until it breaks or the gains run out | Rarely before Series B |
| Works at rate, at margin | Install time, uptime and gross margin per unit moving the right way | Rarely before scale |
| Keeps working | Third-year uptime, service cost per unit, spares lead times | After the money has gone in |
What the capital thinks it is paying for
A bridge
What it is actually paying for
The valley of deathThe same bridge
Two sites prove little on their own. What counts is the nth: whether the run holds site after site, whether each one still comes in quicker and cheaper than the last or the gains have started to run out, and where the run breaks.
Valuations in this sector are routinely set at the fourth rung on evidence from the first. The distance between the two is what the capital is paying for, whether or not the term sheet says so.
The expert in the room
The usual remedy is to put someone with twenty years in automation in front of the deal. That helps. An experienced engineer will see in an afternoon what a well-made deck conceals: the integration nobody has costed, the service model that does not exist yet, the payback calculation that assumes the customer’s line stands idle while the robot is fitted.
Experience has limits too. Rethink Robotics was founded in 2008 by Rodney Brooks, a co-founder of iRobot, and raised close to $150 million. Its robots, Baxter and Sawyer, were widely admired. The original company closed in October 2018, after sales fell short of expectations and an expected acquisition fell through. Pedigree in the room did not supply the later rungs.
Screening has a second limit. A refusal protects the investor from deals that should not be done. It says little about the deals that clear the bar and then spend two years and another round learning how to ship.
The question
The better question is asked before the cheque and again after it: what must this business prove it can do for the nth consecutive time, at rate, on a customer’s floor, and what evidence, by what date, would show it has?
The question can be answered in numbers: install hours, uptime, cost to serve, months of runway. It applies equally to the deal that should be declined, the deal that should be backed, and the company that has already raised and is finding the ladder steeper than its plan assumed. It also turns a verdict into a programme of work, setting out which rungs are missing, who has to learn to climb them, and what that will cost in money and time.
If the board cannot answer it, the valuation is pricing a demonstration.
The missing discipline
Climbing the ladder is engineering work of a particular kind. Manufacturing Engineers design installations that survive a customer’s floor. They qualify suppliers and spares before the fleet needs them. They set service intervals and measure what each site costs to keep running. They write down how the machine goes in, so that the second site is quicker than the first and the tenth quicker again.
Kaipability works between the demonstration and the fleet. It establishes, on evidence, which rungs a company has reached and what it will take to reach the rest. Investors and founders weighing that distance are welcome to start the conversation.
If you are weighing a Physical AI company somewhere between the demonstration and the fleet, that is the conversation to have.
Start a conversation How we assess it →Questions this dispatch answers.
Written to be quoted by AI assistants and search engines. Self-contained answers, verdict first.
