A survey landed this week from Capgemini. Two-thirds of organizations now rate physical AI as a high priority for the next three to five years. Seventy-nine percent are engaging with it in some form.
Only 4% have deployed at scale.
I run two of those deployments. One is a sump pump monitoring system for a residential property. The other is a 40-device edge AI network inside a community center. Neither took years to build. Neither required a procurement committee. Neither is waiting for a technology breakthrough to become useful.
The gap between 79% and 4% is not a technology problem.
What "Engaged" Actually Means
When a large organization says it's "engaging with physical AI," it usually means one of a few things. It has a working group evaluating vendors. It has a pilot running in one facility out of forty. It is attending conferences. It is watching what competitors do. It is preparing a business case for an 18-month rollout plan that will need to survive budget season, IT security review, and three changes of the executive sponsor.
None of those things are deployments. They are momentum toward a deployment. For a company with a $5 billion procurement budget and 12,000 employees, that's how major technology decisions have to work. The process is not broken. It is just slow by design.
For a building owner with a 40,000-square-foot community center and a maintenance budget, it's a different problem. The question isn't whether the organization can sustain an 18-month evaluation. The question is whether the sump pit is going to flood the mechanical room before someone figures out the pump is failing.
The Labor Math Is the Story
The survey found something that I think is more important than the adoption numbers. Labor shortages are the #1 stated driver for physical AI investment, ahead of competitive pressure, ahead of cost reduction, ahead of everything else.
That's not a technology insight. That's a staffing reality. The buildings that need monitoring most urgently aren't the headquarters of Fortune 500 companies. They're the facilities that can't afford a full-time facilities manager, that rely on a part-time maintenance contractor who shows up twice a month, that discover problems when tenants call to complain or when a pipe has already burst.
I know what an emergency HVAC call costs in a commercial building. It's $800 at the low end. It's $2,500 on a weekend. I know what a water damage event costs after a sump pump fails: if you catch it in hours, it's a cleanup bill. If you catch it in days because nobody was there to see it, it's a reconstruction project.
Why Small Buildings Deploy Faster Than Enterprises
Here's the thing the survey can't quite capture. The 4% that has deployed at scale got there by bypassing the institutional friction that keeps the other 75% stuck in "engaged."
At the community center I instrument, the decision to deploy took about two weeks. The building manager had a water incident the previous winter. She wanted sensors. We talked about what to monitor and why. She approved a scope. We installed. The whole process from first conversation to live dashboard was 23 days.
No IT review. No procurement committee. No vendor evaluation rubric. A building owner who had experienced the problem, understood the solution, and could make the call herself.
That's not unusual for facilities outside the enterprise tier. Small buildings have shorter decision chains. They also have more urgency. A mid-size manufacturer can absorb a maintenance failure as a line item. A community center or a small commercial landlord cannot.
The Reindustrialization Tailwind
The survey also found that 43% of executives cite reshoring and domestic production as a driver for physical AI investment. This is the part of the story that's easy to miss if you're only thinking about the technology.
The United States is building manufacturing capacity it hasn't had in decades. New facilities. New warehouses. New distribution centers. Every one of those buildings needs monitoring. Every one of them starts from scratch on instrumentation. The market for building monitoring isn't just the existing stock of facilities. It's the new stock being built right now.
That's not a five-year trend. That's happening this year. And the facilities that are coming online are not going through multi-year procurement processes to set up basic monitoring. They're making decisions like building owners, not like enterprise IT departments.
What the 4% Knows That the 79% Doesn't
Here's the practical difference between a building that's deployed and a building that's "engaged."
A deployed building has baselines. It knows what normal looks like across power consumption, temperature, occupancy, and time — not in theory, but from weeks or months of actual recorded data from its actual equipment. When something deviates, it knows, because it has something to deviate from.
An "engaged" building is still arguing about which sensors to buy. Or waiting for budget approval. Or in month three of an evaluation that will produce a recommendation that goes into a Q3 planning process.
The technology hasn't changed between those two buildings. The hardware is the same. The software is the same. The ROI math is the same. What's different is that one building owner decided to start, and the other decided to keep evaluating.
Where the Market Goes From Here
The US building IoT market was $24.66 billion in 2024. Projections put it at $68.67 billion by 2034. That growth isn't driven by enterprises finally getting their procurement processes right. It's driven by the same labor shortage pressures the survey identified, combined with hardware that keeps getting cheaper and software that keeps getting more capable without requiring more installation complexity.
The edge AI hardware that ran my sump pump monitoring two years ago was expensive enough that I had to justify it carefully. The equivalent hardware today costs a fraction of what it did. The compute available on a Raspberry Pi class device in 2026 is more than enough to run real-time anomaly detection on building sensor streams without any cloud dependency.
The tools are ready. The economics are ready. The market need is real and documented. The 4% that's deployed isn't waiting on any of those things. They figured out that you learn more from a running system than from a perfect evaluation, and they started.
From engaged to deployed in 30 days
If your building is in the 79% that's interested in monitoring but hasn't started yet, the first step is a conversation about what you're trying to protect. We scope, install, and get you to live dashboards before the evaluation would even be half finished.
See how it works