A market report landed this week that put a number on the thing I do for a living. Physical AI, machines that sense the world, decide, and act, is now pegged at over $430 billion by 2030, on its way to roughly $1.6 trillion by 2040. The sector pulled in more than $75 billion of investment last year alone. Japan by itself is putting $65 billion behind it.
Big numbers. But the useful part of a market report is never the total. It is the taxonomy, the way the analysts carve the pie, because that tells you where the money is actually pointed. And this one carved the whole $430 billion into exactly nine vertical sectors.
Here they are: industrial automation, autonomous vehicles, robots, smart infrastructure, healthcare, agritech, defence, space, and consumer systems.
Read that list again with a specific building in mind. A church with a boiler in the basement. A three-unit rental with a sump pump that runs every spring. A small machine shop with one aging compressor. Now find the vertical.
There isn't one
A church basement is not "smart infrastructure." That vertical means power grids, highways, water utilities, the stuff cities and states buy. A small rental is not "industrial automation," which means factories with maintenance engineers on the payroll and a plant manager who signs seven-figure POs. A machine shop with one compressor is not "robots," and it is very much not "defence" or "space."
The small commercial building falls straight through the seams of all nine. Even where the report does mention predictive maintenance and smart buildings, it files them under industrial automation and frames them at enterprise scale. The building I work on, the one with nobody in the basement and no engineer on call, isn't small enough to be a footnote. It just isn't there.
And that is not a sad footnote. That gap is the entire business.
Why the seam is the opportunity
Every one of those nine verticals is being chased by companies that need enterprise-scale deals to justify their burn. Their products get priced, packaged, and staffed accordingly: integration teams, six-figure deployments, custom-silicon roadmaps, a sales motion that only pencils out above a certain deal size. That machinery cannot physically bend down to a $99-a-month church basement. The unit economics forbid it.
So the building with real equipment, real failure modes, and a real 2 a.m. flood risk gets served by nobody, precisely because it is too small to be a vertical and too real to not have the problem.
Here is the part that makes it work. The same $430 billion that is funding the nine verticals I'll never compete in is also, as a side effect, mass-producing the components I build from. Vibration sensors that cost $600 a few years ago cost $50 now. Small models that beat much larger ones on one narrow physical task. Edge boxes that run inference locally for the price of a doorbell. I don't have to fund a vertical. I get to shop in the clearance aisle of nine of them.
The frontier is spending billions to build a machine that can handle a task it has never seen, in an environment it has never been. My monitor has the friendly version of that problem: one machine, one job, a basement that is either dry or not every single morning.
The Watertown pump was never on the map, and never needed to be
The sump pump I've been monitoring in Watertown does not appear in any market report. It is not smart infrastructure. It is a pump in a basement. But it has a bearing that will someday fail, a motor that draws more current as it strains, and a vibration signature that starts drifting weeks before anything goes wrong. All the physics that a $430 billion market is chasing at the frontier is already sitting in that pump, at a scale one person with off-the-shelf parts can actually serve.
Same story at the 40-device building in Northampton. Nobody would call it a "vertical." It is a normal small commercial building. But it generates its own operating data every day, it has assets that degrade in readable steps, and it has nobody standing in the mechanical room to notice. A monitor is the only thing between a slow degradation and an expensive surprise. That job does not need a world model or a humanoid. It needs a $50 sensor, a small model on a local device, and someone who bothered to show up for a building that isn't on anyone's map.
| The nine-vertical frontier | A small-building monitor |
|---|---|
| Must generalize across tasks and environments it has never seen | One machine, one job, graded by reality every morning |
| Enterprise deal sizes, integration teams, custom silicon | Off-the-shelf sensor, small local model, clamps on in an afternoon |
| Priced for factories, fleets, grids, hospitals | $99–$199/mo, hardware under $3,000 |
| Data leaves the site for cloud training and inference | Nothing leaves the building |
The pitch is not "we do physical AI too"
When a market gets this loud, the temptation for a small operator is to stand next to it and say "we do that too." That's the wrong pitch, because I'm not doing what the nine verticals do, and I'd lose that fight on scale every time.
The right pitch is the opposite. I serve the building none of that money is priced for. The pump, the boiler, the compressor in a place with nobody in the basement. I'll clamp a detector on your most critical asset, it learns that machine's normal in a couple of weeks, and it raises its hand three to six weeks before a failure, using the same physics the frontier is spending billions to reach, at a price a small building can actually pay. The $430 billion market and I are not competing. We're barely in the same room. It builds the frontier. I show up for the basement.
Your building isn't a vertical. It still has a pump worth watching.
Each critical asset, your pump, your boiler, your compressor, gets its own small detector built from off-the-shelf sensors, running a small model on a local device. It clamps on in an afternoon, learns your machine's normal, and watches vibration, temperature, and current for the drift that shows up weeks before a breakdown. Then it tells a person who knows the equipment. Nothing leaves the building. $99 to $199 per month, hardware under $3,000.
See how it worksSources: GlobeNewswire, "Physical AI Market Set to Surpass $430 Billion by 2030, Driven by Nine Key Vertical Sectors," July 1 2026 (nine verticals: industrial automation, autonomous vehicles, robots, smart infrastructure, healthcare, agritech, defence, space, consumer systems; $430B by 2030, ~$1.6T by 2040; $75B+ raised in 2025; Japan's $65B robotics commitment). NVIDIA, "National Robotics Week 2026" (Isaac GR00T, Cosmos world models, Newton 1.0), for the frontier-capability context. Predictive-maintenance market $9.71B (2026) to $16.74B (2031) at 11.5% CAGR, MarketsandMarkets. Edge AI market and NPU efficiency figures from 2026 industry coverage. Field deployments at The Intersecto Watertown sump-pump site and Northampton 40-device building. Companion brief: /Users/tdeshane/lobster/research/physical-ai-brief-2026-07-07.md.