People keep asking me what got smart enough that I can watch a single machine in a single basement and have it pay for itself. They assume the answer is the AI. It isn't. The anomaly model on my sump pump in Watertown is not exotic, and it has not gotten dramatically smarter in two years. The thing that changed is duller and far more important: the sensors got cheap.
Here is the number that crossed my desk this week. Industrial IoT sensor hardware has fallen about 85% since 2019. A vibration monitoring node that cost $600 per point six years ago now costs under $50. And the consequence of that one price move is the entire reason this is a business at all.
The ROI line used to live in a factory
Condition monitoring was a big-industry product for a simple reason: the sensors were a big-industry expense. When a single vibration point cost $600, you only bolted them onto machines whose failure cost six figures, the turbine, the production line, the thing that shuts down a plant. The math never closed on anything smaller. A boiler, a rooftop HVAC unit, an elevator motor, a pump in a finished basement were all below the line. Not because they didn't matter. Because the instrumentation cost more than the worry.
That line just moved. At under $50 a node, the trade press now puts positive ROI on essentially any asset worth more than about $5,000. Sit with how far that drops the threshold. It isn't "factories" anymore. It's every building with a mechanical room. The same six sensor types that used to be reserved for industrial plants, vibration, temperature, current, acoustic, pressure, humidity, now pencil out on a single commercial boiler.
Cheap sensing is half. Local inference is the other half.
A cheap sensor that ships its data to a cloud you rent by the call is not actually cheap, because you have traded a one-time hardware cost for a forever metering bill. The second thing that had to happen, and has, is that the intelligence moved next to the machine. The industry reporting this week says the same thing in its own words: anomaly detection is moving off the cloud and onto the edge, processed locally at sub-100ms latency, with the cloud demoted to long-term storage. Failure rates down as much as 50%, maintenance costs down as much as 40%, and the decision happening right where the machine lives.
Put the two together and the stack for watching one machine is small, finished, and affordable in a way it simply was not in 2019. A handful of sub-$50 sensors. A low-power board on the wall. A narrow anomaly model that learns one machine's normal and watches it many times a minute, forever, with no round-trip and no per-call meter. Under $3,000 of hardware, total.
| 2019 | 2026 |
|---|---|
| Vibration point: ~$600 | Vibration point: under $50 |
| ROI only on six-figure-failure assets | ROI on any asset worth more than ~$5,000 |
| Inference in the cloud, per-call billing | Inference on the edge, sub-100ms, no meter |
| Condition monitoring is a factory product | Condition monitoring fits a single basement |
What this looks like on an actual machine
My sump pump in Watertown is the limit case of this. It is not a turbine. By the old math it was never worth instrumenting, the sensors would have cost more than the pump. By the new math it is obvious: a few cheap sensors reading vibration and current, a small board that learned what a healthy cycle sounds like, and an alert the moment the cycle drifts. The asset it protects, a finished basement, is worth far more than the $5,000 ROI threshold, and the hardware watching it costs less than a single emergency call-out.
The forty-device building in Northampton is the same idea repeated. Not one giant brain watching everything from a datacenter, but a cheap, local, narrow watcher per critical machine, each one finally cheap enough to justify on its own. That was a fantasy at $600 a point. It's a purchase order at $50.
What this means if you own a building
If you ever priced out condition monitoring and walked away because it only made sense for industrial equipment you don't own, the answer changed underneath you. The thing that priced you out, the cost of the sensors, fell by 85%. The ROI line dropped from the factory floor down to the mechanical room in your own building. Your boiler, your rooftop unit, your elevator motor, your pump each produce a small, steady signal about their own health, and for the first time it costs less to listen to that signal than to be surprised by its absence. You don't need a smarter model. You need cheap sensors and a watcher on the wall, and both are finally here.
Cheap sensors, a watcher on the wall, one machine at a time.
Each critical asset gets its own small detector, built from off-the-shelf sensors that cost a fraction of what they did a few years ago, trained on its own native signals, running on a low-power board bolted next to the machine, local and watching 24/7. No round-trips, no per-call billing, no data leaving the building. It learns the machine's normal and warns you early when it drifts. $99 to $199 per month, hardware under $3,000.
See how it worksSources: Industrial IoT sensor cost decline (~85% since 2019; vibration monitoring node from ~$600 per point in 2019 to under $50 in 2026; positive ROI now on assets valued above ~$5,000 across all six major sensor types) from the oxmaint 2026 IoT-sensors-for-predictive-maintenance guide. Edge-AI architecture shift (local anomaly detection at sub-100ms latency, cloud demoted to aggregation/storage, failure rates cut up to 50% and maintenance costs up to 40%) from EE Times, "Edge AI Is Forcing a Rethink of Predictive Maintenance Architecture," and related June 2026 trade reporting. Predictive-maintenance market context (~$9.71B in 2026, edge AI the fastest-growing segment at ~14.2%) from MarketsandMarkets. Field deployments at The Intersecto Watertown sump-pump site and Northampton 40-device building. Companion brief: /Users/tdeshane/lobster/research/physical-ai-brief-2026-06-30.md.