The Humanoids Are Sold Out. Your Building's Parts Are On a Shelf.

Two numbers crossed my desk this week, and together they tell you almost everything about where physical AI actually is right now.

The first: Figure's robot factory, BotQ, hit a rate of one humanoid robot per hour. That's a twenty-four-fold jump from one a day, in under four months. They've delivered a few hundred so far, with end-of-line yield north of eighty percent, and they're treating the ramp as their single most important project.

The second: Boston Dynamics' entire 2026 production run of the new electric Atlas is already gone. Every unit they build this year is committed to exactly two customers, its parent company Hyundai and Google DeepMind. You cannot buy one. Not for a premium, not with a wait. The whole year is allocated. Hyundai is so sure robots will stay scarce that it's putting twenty-six billion dollars into U.S. operations, part of it a new factory designed to make thirty thousand robots a year.

I build physical AI for small buildings. A sump pump in a basement in Watertown. Forty devices in a building in Northampton. And when I read those two stories back to back, the thing that struck me wasn't the speed or the money. It was the scarcity. The frontier robot has become a rationed good. And that fact, more than any demo, is what makes building owners think they've already missed the boat.

The robot is now a supply problem

Step back and look at what those numbers are really saying. Figure is proud that it can make one robot an hour, and it's racing to make that number bigger, because the number is the bottleneck. Boston Dynamics' whole year is spoken for before a single unit reaches an outside buyer. Hyundai is spending the GDP of a small country to build capacity. Every one of those facts points the same direction: the frontier robot is scarce. There is a yield curve. There is a waitlist. There is an allocation queue, and it runs through the largest companies on earth.

That's not a criticism. It's exactly what you'd expect from a genuinely hard, genuinely new machine. A general-purpose humanoid that can walk, balance, manipulate, and learn is one of the hardest things anyone has ever tried to mass-produce. Of course it's rationed in 2026. Of course the first units go to a carmaker and an AI lab. That is what the early years of a hard hardware product look like.

The thing that protects your building has the opposite supply curve

Now look at what actually keeps a small building from getting wrecked by a failure. It isn't a humanoid. It's a vibration sensor on a pump. A current clamp on a motor. A temperature probe on a bearing. A small edge box on the wall running a model that learned what that one machine sounds like when it's healthy.

And every single one of those parts is a commodity that's in stock today. No waitlist. No allocation to a car company. No yield ramp standing between you and the hardware. These parts came off production lines that were finished and paid for years ago, amortized across millions of units in consumer electronics and industrial IoT. You can put the whole kit in a cart this afternoon and have it bolted to a joist by Friday.

The trap in "humanoids shift to production"

Here's the conclusion a building owner reaches from a week of headlines like these. They read that humanoids are "shifting from pilot to production," that Atlas is sold out for the year, that Hyundai is spending twenty-six billion dollars to build more. And they think: physical AI is being rationed to the giants. The hardware all goes to Amazon and Hyundai and DeepMind first. I'm late, I'm small, I'm at the back of a line I can't even see the front of.

That's backwards, and the supply numbers are exactly why. The part of physical AI that's scarce, waitlisted, and allocated to hyperscalers is the robot, the part that's genuinely hard to build and genuinely in short supply. The part that watches your pump was never the scarce part. It's commodity silicon that's been in volume production for a decade. There is no allocation queue for a fifty-dollar vibration sensor. You are not behind Hyundai in line, because the thing that protects your building was never on the same shelf as the thing Hyundai is buying.

What "in stock" actually looks like

The sump pump in that Watertown basement is the whole argument in one object. Nothing about it was scarce. The vibration sensor, the temperature sensor, the current clamp, the little edge box that runs the detector on the wall, all of it was ordinary, available, off-the-shelf hardware I could order without joining a waitlist or outbidding a car company. I screwed the sensor to the joist, let the model learn what that specific pump sounds like on that specific duty cycle, and from that day on its only job has been to notice when that normal slips.

There was no allocation queue. No first-pass yield to wait out. No factory being built in Georgia so that someday I might get my turn. The watchman's "production problem" was solved a decade ago, by the same supply chains that put a microcontroller in every thermostat and a MEMS accelerometer in every phone. The frontier is still figuring out how to build its robots fast enough. The building owner's hardware has been sitting in stock the entire time.

And the in-stock part got cheap, too

Scarcity usually means expensive. The frontier robot is both scarce and costly, which is the worst combination if you're a small operator. The watchman is the reverse, abundant and cheap. A vibration-monitoring node that cost around six hundred dollars a point in 2019 is under fifty dollars in 2026. Wiring up twenty points in a mechanical room is roughly nine hundred dollars in sensors. That price crossed a line most owners haven't noticed: it now pays for itself on any asset worth five thousand dollars or more, which describes nearly every pump, boiler, compressor, and rooftop unit in a small commercial building.

So the comparison lands like this. Boston Dynamics will spend the year building robots it has already promised to two of the biggest companies in the world. You can put a watchman on the asset that would actually hurt if it failed, this afternoon, with parts that are in stock right now, for less than dinner for two, and get something that warns you weeks before that asset dies. Different machine, different supply curve, different price. The one that protects your building is the abundant one.

What this means if you own a building, not a warehouse

When the whole industry spends a week talking about humanoids going into production and the entire year's output selling out before launch, it's easy to feel like physical AI has been handed out to the giants and there's nothing left for you. For the robots, that's literally true this year. But that scarcity is specific to the robot. It is not the scarcity of watching your pump, because watching your pump was never scarce.

Your building isn't a warehouse and it never will be. It doesn't need a humanoid, and it couldn't get one this year if it did. What it has is a handful of assets that have been quietly telling the story of their own health the whole time, with nobody listening, and the hardware to start listening is sitting in stock, cheap, today. Pick the few assets that would actually hurt if they failed, the pump, the boiler, the compressor, the twenty percent that cause eighty percent of your trouble, and give each one a watchman. No waitlist. No allocation. No turn to wait for. Just commodity parts, bolted in place, learning your machine from your machine, starting this week.

Sources: Figure AI's BotQ facility crossing one Figure 03 per hour, a 24x throughput increase in under 120 days (350+ delivered, 500+ shipped, 9,000+ actuators, end-of-line first-pass yield >80%, battery yield 99.3%, 80+ functional tests/unit), Figure framing the ramp as a development-velocity milestone (Figure AI / Interesting Engineering / Humanoids Daily, May–June 2026); Boston Dynamics' entire 2026 electric Atlas production run committed to Hyundai (RMAC) and Google DeepMind, with Hyundai (full owner after SoftBank's $325M exit) announcing a $26B U.S. investment incl. a robotics factory targeting 30,000 robots/year and Atlas deployment at the Savannah, GA Metaplant by 2028 (Boston Dynamics / Hyundai Newsroom, 2026); NVIDIA Cosmos 3 world model trained on ~20 trillion tokens, Jim Fan / GEAR Lab "year of world models," DreamDojo checkpoint ranking (NVIDIA / Axios, June 2026); Automate 2026 Humanoid Robot Forum (Chicago, June 23–24); Weekly Robotics #365; condition-monitoring node price falling from ~$600/point (2019) to under $50 (2026), crossing positive-ROI on assets worth $5,000+; edge AI market $24.91B (2025) projected to $118.69B (2033) at ~21.7% CAGR (Grand View Research). Field deployments at The Intersecto Watertown sump-pump site and Northampton 40-device building. Companion brief: /Users/tdeshane/lobster/research/physical-ai-brief-2026-06-24.md.