A mid-sized plant manager reads the same robotics headlines you do. Physical AI raising billions, humanoids on every trade-show stage, a foundation model that supposedly generalizes across any task on the floor. Then he walks his own line, where scheduling still runs on a spreadsheet and not one machine moves on its own. The distance between what the vendors announce and what the floor actually runs is the defining tension in manufacturing right now, and a number back in circulation this month puts a hard edge on it. Eighty percent of United States manufacturing facilities operate with zero automation. Not lagging, not partial. Zero. For a hardware founder selling into that market, that is uncomfortable and useful at the same time: the capability race the industry keeps covering is not the race that decides who wins.
Belief is high. Deployment is almost nothing.
The figure comes from Intrinsic chief technology officer Brian Gerkey, whose company builds robotics software inside Google, and was reported by Manufacturing Dive. Set it against how manufacturers talk about their own future. Deloitte’s 2025 Smart Manufacturing survey found that ninety-two percent believe smart manufacturing will be the primary driver of competitiveness over the next three years. Yet only about twenty-nine percent said they already use AI or machine learning at the facility or network level, and just twenty-four percent have deployed generative AI. Jeff Burnstein, who runs the Association for Advancing Automation, put the gap plainly to Manufacturing Dive: interest is high across the board, but execution is where things get difficult. Belief is nearly universal. Deployment is a rounding error against it.
The bottleneck is not the robot
When a project stalls, founders assume the machine was not good enough. On a factory floor it is almost never the machine. The barriers manufacturers actually name are cost, integration complexity, a shortage of people who can deploy and maintain the system, and no clear return. Jasmeet Singh of Infosys described plants still running on fragmented legacy systems, with data that was never structured for a robot to use, and buyers who want proven payback before they commit real money. The hard step is not the demo. It is the jump from a pilot to a measurable business outcome, and that is exactly where most of these projects die.
I have watched this failure mode well outside manufacturing, and it always rhymes. Physical AI has its own version, the graveyard of pizza-delivery robots and their cousins: impressive machines, real working demos, dead companies. They fell in love with their engineering, not with their customer. A working demo feels like progress, but it is not progress if nobody has proven they will pay for it at a margin that lets the company live. The plant manager is not buying a robot. He is buying lower cost or lower liability on one specific process, and if the machine does not deliver that, what are you actually selling him?
The economics have to close on the first line, not the fiftieth
This is where a lot of robotics founders get the sequence backward. Unit economics do not gate your hundredth deployment. They gate your first one. A plant will not fund a science experiment, so the payback math has to make sense on day one or you never get onto the floor at all. That discipline is why the more interesting news this month is not another billion-dollar physical-AI round. It is a wave of vendors meeting the eighty percent much closer to zero. No-code cobots that need no dedicated cell and no specialist programmer, from Hirebotics on an explosion-proof painter to Productive Robotics on machining. Ambi Robotics and Pickle Robot shipping what they call the first commercial physical-AI system that unloads a truck and palletizes with no person in the loop. Techman Robot naming the culprit directly, the hidden cost trap of long commissioning cycles and endless reprogramming that quietly erodes the payback. Every one of those moves lowers the cost of getting started, which is the number the buyer is actually weighing.
Dave’s take
Eighty percent of factories sitting at zero is not a story about robots that cannot do the work. It is a market waiting for someone to make the deployment and the day-one math close. When a physical-AI founder walks me through how the model benchmarks, I care less about that and more about whether a plant manager can run the payback in his head and say yes on the first line. Build the customer proof before you build the fleet. The companies that win this cycle will treat integration cost and unit economics as product decisions, not as someone else’s problem to solve after the demo lands.
From Dave’s video library
Dave on separating where automation actually returns money from where you are just feeding the meter, the same payback question a plant manager asks before a robot ever reaches the floor.
I’m here to help you scale.
Work With DavePrefer a smaller first step? Book a $500 one-hour working session →
Dave Saunders is the founder of Base Reality Group and a Fractional CPO for hard-tech founders. He was a founder and operator at Galen Robotics, where the surgical-robotics platform earned FDA De Novo authorization in 2023, and he managed a 35-patent portfolio licensed from Johns Hopkins. He wrote Founders Who Finish and publishes The Build. More about Dave →