The biggest artificial intelligence conference in the world ran this week in Shanghai. More than 1,100 exhibitors. Over 3,000 exhibits. More than 300 products making their global debut, which is to say, appearing in public for the first time anywhere on earth. Humanoids that walk, humanoids that fold laundry, a robot hand with more fingers' worth of engineering in it than most cars have in the whole steering column.
I read through the coverage the way I read all of it, which is with one specific question in mind: what does this cost and when can I have it? And that question kept coming back empty. Three hundred debuts. Almost none of them with a price. Almost none of them with an order page. You cannot buy nearly anything that was unveiled at the largest robotics showcase of 2026.
I want to be careful here, because that is not a criticism of the engineering. The engineering is genuinely astonishing. But a missing price tag is a fact about a product, and it happens to be the most useful fact I learned this week.
The one number that was published
Unitree did put a number on something. They showed the GD01, which they describe as the world's first mass-produced transformable robot: two point seven meters tall, five hundred kilograms, able to switch between walking on two legs and four, rated to carry a person. It starts at 3.9 million yuan. That's about $540,000.
It's a magnificent machine and I am glad it exists. It is also, for the roughly one hundred percent of small building owners reading this, a number from a different planet. The only humanoid you can actually check out with today is the 1X NEO at $20,000 to own or $499 a month, and its first home deliveries are still described as "later in 2026."
Boston Dynamics is the part of the story nobody wants to talk about
The other thing that happened this week: Hyundai confirmed it is buying SoftBank's remaining stake in Boston Dynamics, roughly ten percent, for about $325 million. Boston Dynamics is now wholly owned by a car company.
The mechanism is what caught me. When Hyundai bought in back in 2021, SoftBank kept a put option, a contractual right to sell its shares back if Boston Dynamics had not gone public within five years. The deadline was June of this year. It came and went. So SoftBank exercised, and Hyundai wrote the check.
Think about what that means. Boston Dynamics is thirty-four years old. It has built the most advanced legged robots any human has ever seen. Its videos have been watched hundreds of millions of times. And in five years it could not reach a public listing, and its most famous machine still has no price and no order page.
I say that with real respect. General-purpose physical capability, a machine that can do anything in a space it has never been in, is one of the hardest problems anyone has ever taken on. It is a decades-long capital sink. Roughly half a billion dollars went into humanoid startups just this week: a $300 million seed round at a billion-dollar valuation for one company, a $200 million pre-IPO round for another, and a Chinese listing that as of Monday still had no share price. That capital is real and the ambition is real.
But none of it is coming for the pump in your basement. Not this year, not next year.
What I do instead, and what it costs
I put one off-the-shelf sensor on a piece of equipment that matters. A pump, a compressor, a boiler, a rooftop unit. It runs into a small local box with a small model on it. Over the first couple of weeks that box learns what your specific machine's normal sounds and feels like: the vibration signature it always has, the current it pulls on a hot day versus a cold one, the way it starts and the way it settles. Then its whole job, forever, is to notice when today stops matching yesterday.
That's it. That is the entire product. It never walks anywhere. It has no arms. It cannot fold laundry or carry a person or transform between two legs and four. It sits bolted to one machine and pays attention.
It costs $99 to $199 a month, hardware under $3,000, and it installs in an afternoon. Nothing leaves the building. There is no cloud subscription underneath it and no per-query API bill, because the model is small enough to run on a cheap board that I hand you along with the sensor.
"Shouldn't I just wait for the robots?"
I get asked this, in some form, on most calls. It is a completely reasonable question. The honest answer got sharper this week.
Waiting is a bet that general-purpose machines will arrive soon enough and cheap enough to solve a specific problem you have right now. This week that bet got a price check: the flagship product from the flagship company is half a million dollars with no delivery date, the most famous robotics firm on earth just got taken fully private by an automaker after missing its IPO window, and the money pouring in is going toward capability, not toward affordability in your boiler room.
Meanwhile the specific problem is not waiting. Bearings degrade on their own schedule. A sump pump in Watertown that I instrumented over a year ago has been running the whole time on a sensor and a small local model, and the failure it is watching for does not care what got announced in Shanghai. Neither does the forty-device building in Northampton. Those systems were cheap when I built them and they got cheaper since, because the component curve keeps falling: a vibration sensor that cost $600 a few years ago is $50 now, and the newest ones run the first pass of anomaly detection inside the $25 part itself.
That last bit is the actual relationship between the frontier and my end of the business. All that humanoid capital buys actuator volume, sensor volume, and small-model research. Every dollar of it quietly makes the boring monitor cheaper. I do not compete with the robots. I get paid, indirectly, by the fact that they exist.
What to take from this if you own a building, not a robot
- A missing price is a real signal. Three hundred global debuts and almost no order pages means the category is still upstream of deployment. Impressive is not the same as available.
- Capital intensity tells you the timeline. Thirty-four years and a missed IPO deadline at the best-known company in robotics is not a knock on them; it is a measure of how long the general-purpose problem takes. Narrow problems get solved and priced first.
- The narrow version already works and already costs something specific. One sensor, one machine, one local box, two weeks to a baseline, three to six weeks of warning before a failure, $99 to $199 a month. You do not have to wait for any of it.
Last week I wrote about a frontier lab teaching a robot to cross an unfamiliar building with one camera and no map, and why my monitor wins by never moving. This week is the commercial version of the same point. The hard, expensive, decades-long work is happening in Shanghai and Waltham and it is worth every dollar. The easy, cheap, already-solved work is available now, has a number next to it, and stops your floor from flooding. I sell the second one on purpose.
Your equipment does not need a robot. It needs something paying attention.
Each critical asset, your compressor, your boiler, your pump, gets one off-the-shelf sensor and a small model on a local box. It learns that machine's normal over the first couple of weeks, then watches for the day it drifts, catches trouble three to six weeks before a failure, and hands it to a person who knows the equipment. Nothing leaves the building. $99 to $199 per month, hardware under $3,000, installed this week.
See how it worksSources: World Artificial Intelligence Conference 2026, Shanghai, July 17–20 (1,100+ exhibitors, 3,000+ exhibits, 300+ global product debuts), via Global Times and CGTN. AgiBot A3 Ultra, X2 Edu, G2 Max and OmniHand 3 Ultra-M debuts, July 18, via TechTimes. Unitree GD01 transformable robot (2.7m, 500kg, biped/quadruped, rated to carry a person; from ¥3.9 million, approx. $540,000) and the absence of public prices or order pages on WAIC humanoid debuts, via RoboZaps weekly roundup, July 13–20 2026. 1X NEO consumer pricing ($20,000 to own or $499/month; first home deliveries promised later in 2026). Hyundai Motor Group statement on Boston Dynamics, July 16 2026 (acquisition of SoftBank's remaining approx. 10% stake, reported at approx. $325 million, following SoftBank's exercise of a put option contingent on no public listing by June 2026), via Hyundai Newsroom, Bloomberg and Reuters. Humanoid funding: Walden Robotics $300M seed at $1.1B valuation, July 15; LimX Dynamics approx. $200M pre-IPO at approx. $2.21B; Unitree STAR Market listing registration approved July 3, still unpriced as of July 20, via CNBC and RoboZaps. EngineAI Ultimate Robot Knock-out Legend, Shenzhen, opened July 17 (32 teams, standardized T800 platforms, ¥10 million prize pool). Condition-monitoring hardware context: STMicroelectronics IIS3DWB10IS in-sensor-AI vibration sensor (ISPU 2.0), from $25 in 1,000-piece orders, available July 2026, via Edge AI and Vision Alliance. Edge predictive-maintenance economics (local inference <200ms; 50–70% reduction in unplanned outages within the first quarter; 12–18% lower overall maintenance cost; 20–40% asset-life extension) via IndustryWeek and EE Times, 2026. Field deployments at The Intersecto Watertown sump-pump site and Northampton 40-device building. Companion brief: /Users/tdeshane/lobster/research/physical-ai-brief-2026-07-20.md.