In a single week, three different markets quietly crossed the same line.
Locus Robotics, the company most warehouses already know for autonomous mobile robots, launched the Locus Array — a fully autonomous picker that combines an omnidirectional mobile base, a vision system, and a robot arm into one product you can buy. Skild, the foundation-model startup that's been promising general-purpose robot brains, picked up Fetch Robotics' hardware and customer base out of Zebra. And the California DMV opened the entire state to Class 8 autonomous trucks, removing the last regulatory ceiling on driverless freight in the country's biggest logistics market.
Three different verticals — warehouse picking, robot brains, autonomous trucking. Three different buyers. Three different timelines. Same week. And in each case, the same pattern: capability has been there for a while, the economics finally penciled, and a regulatory or distribution window opened.
That pattern has a name in deployment circles. It's what physical AI looks like leaving the demo phase.
The Three-Step Pattern
Every physical AI category I've watched make this transition has followed roughly the same sequence.
Step one: capability proven. Someone publishes a paper, demos a prototype, or runs a high-profile pilot. Sony AI's Project Ace, which formally published in Nature the same week Locus shipped Array, is this step at its most theatrical — a robot beating elite and professional table tennis players in matches under official ITTF rules, using nine high-speed cameras and three event-based vision sensors to track a spinning ball. Most categories don't get a Nature paper. They get a YouTube video and a press release. But the function is the same: capability is no longer the question.
Step two: economics work. Hardware costs come down enough that the deployment math closes for someone with a real budget — usually starting in the highest-cost-of-failure niches. Fetch Robotics existed in the warehouse picker market for years; Skild buying it now is an economics signal, not a capability signal. The hardware was ready. What changed is that a foundation-model company can now justify paying for the customer base because the unit economics of vision-language-action models at warehouse scale finally look investable. Sereact's Series B for Cortex 2.0 the same week, after a billion production picks in Europe, is the same data point from a different angle.
Step three: regulatory or distribution window. The thing that lets the technology scale beyond the early adopters opens up. California's statewide autonomous-truck rule is the cleanest version of this — a single regulatory action that converts a corridor-by-corridor experiment into a deployable product. In other categories the gating step is a major distributor signing on, or a building code update, or the moment when an enterprise buyer's procurement team starts asking for the technology by name.
Once all three preconditions are met simultaneously, the deployment window opens. It typically stays open for 12 to 24 months before consolidation hits — the early operators who shipped during the window keep their customers, and the late entrants either acquire one of the early operators or watch the enterprise vendors move down-market and squeeze the margins.
Where Building Monitoring Sits Right Now
I think building monitoring is in the open window. Here's the evidence.
Capability is proven, on actual buildings, with actual data. I've been running edge AI on two buildings for a while. One is a residential property with a sump pump that nearly burned out from a sticky float switch — the local AI caught the anomaly at 3am during a heavy rain forecast and ran 97 automated recovery cycles. No flood. No motor replacement. The other is a community center with 40 devices across the facility; the AI surfaced an HVAC scheduling pattern that was running on weekends with near-zero occupancy, and fixing it cut energy costs 42%. Neither system needed a foundation model, a cloud subscription, or anything close to the sensor density Sony deployed for table tennis. They needed cheap sensors, a local model, and the discipline to actually look at the data.
The economics work at very small scale. The hardware bill of materials for the residential deployment was under $200. The community center came in under $500. Texas Instruments acquiring Silicon Labs and NVIDIA shipping the Jetson T4000 are pushing the per-sensor capability up while the per-sensor cost stays flat. The IoT For All newsletter I read this week was full of stories about cloud AI subscription pricing breaking — GitHub Copilot moving to usage-based billing, OpenAI projecting an 80% decline in ChatGPT Plus subscriptions for 2026, a Cursor agent powered by Claude deleting a company's entire production database in nine seconds. Buildings have 10 to 30 year operating horizons. They cannot absorb that kind of pricing volatility, and they should not be giving cloud agents write access to their physical infrastructure. Edge architecture is the only honest answer for that horizon, and the unit economics already work.
The regulatory window is essentially open by default. No permits to put a sensor on your own pump. No building code update needed to read your own electrical panel. The constraint is awareness and distribution — the perception that "smart buildings" require a six-figure enterprise deployment, and the absence of any natural channel that introduces the affordable version to the small-to-mid building owner who needs it most.
Two of the three preconditions are met cleanly. The third is a sales and content problem, not a technology problem. The deployment window is open.
What "Deployment Phase" Actually Means for a Practitioner
The window matters because of what comes after it.
When Locus consolidated the warehouse-AMR market starting around 2018, the early operators who deployed during the open window built reference customers, accumulated operational data, and earned the right to be referenced by the next customer. The late entrants — who showed up after capability and economics were already validated — found themselves competing for the customers the early operators hadn't already won, with thinner margins, less differentiated offerings, and an enterprise vendor stack moving down-market behind them.
Building monitoring is on the same curve, just earlier. Right now, the small-to-mid building market is undersupplied. Most owners I talk to fall into one of two buckets: "I didn't know that was something small businesses could do," or "we looked at a building management system years ago and the quote was insane." Both are correct descriptions of the market as it stands today. Neither is a description of the market as it will stand in 2027 or 2028.
By then, the same pattern that played out in warehouses, robot brains, and autonomous trucking — capability + economics + regulatory window — will have run its course in building monitoring. The enterprise vendors will move down-market with packaged offerings. The cloud platforms will offer subscription bundles. The components Sensors Converge is showing this week in Santa Clara will have arrived in the mass-market sensor lineups. The building owners who deployed during the open window will keep the savings; the ones who didn't will be paying enterprise prices to catch up.
What I'm Watching This Week
Sensors Converge runs May 5-7 in Santa Clara, with the Edge AI Foundation pavilion hosting eight component vendors that will be in the next 12-18 months of small-building deployments. The 2026 Robotics Summit & Expo runs May 27-28 in Boston, with a dedicated physical-AI track. Wetour Robotics is launching its Orchestra physical-AI operating system on May 28 in Austin. DARPA's RFI on embedded hardware intelligence for robotics closes May 27 — federal interest at the silicon layer is itself a deployment-phase indicator.
None of these events will mention small-building monitoring. That is itself the opportunity. The same forces that are visibly converging on warehouse robotics, humanoid factories, and autonomous trucking are converging on the building you're sitting in right now. The vendors haven't shown up yet. The window is open.
Get into the deployment window before it closes.
We deploy edge AI monitoring on small-to-mid buildings using off-the-shelf sensors, no enterprise contracts, no cloud lock-in. Under $500 to start. Two case studies running 24/7. We'll tell you what your building is hiding before it becomes an emergency — and we'll do it before the enterprise vendors arrive.
See What We BuildRead the case studies: How edge AI prevented a basement flood | Smart building on a shoestring | The edge AI hardware wave has arrived | DreamDojo doesn't live in your basement