Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.

The Humanoid Robot Revolution: The Body of the New Economy

The Year the Robots Actually Shipped

Remember 2025? That was the year we called the “Model T moment” for humanoid robots, when 1X Neo pre-orders opened at $499/month and Figure 02 helped build 30,000 BMWs. It felt like the dam was breaking, and we had no idea how fast.

2026 is the year the dam broke. The breakthrough showed up in purchase orders, factory deployments, and billion-dollar funding rounds landing every other week rather than in press releases or conference demos. TrendForce projects 50,000+ humanoid robot shipments in 2026, a 700% surge over the previous year — and by June, Morgan Stanley had doubled its China-only forecast twice in one year, from 28,000 units to 50,000, meaning the bank now expects China alone to ship what analysts had projected for the entire planet. The global humanoid robot market is projected at $4-6 billion in 2026, racing toward $38 billion by 2035 and a staggering $5 trillion by 2050.

Then, on July 28, 2026, the US government did something that made the shipment forecasts a geopolitical question rather than an industrial one. More on that below.

In the Unscarcity framework, this is The Body, completing the triad with The Brain (AI cognition) and The Fuel (fusion energy). Together, they form the technological substrate for post-scarcity civilization. The Brain can write your emails and compose your sonnets, but someone still has to take out the trash, build the houses, and care for grandma.

That “someone” increasingly has a serial number.

The Numbers That Changed Everything (Updated)

Let’s revisit the arithmetic that was already uncomfortable in 2025, and got worse.

1X Technologies’ Neo robot: Available at $20,000 outright or $499/month. At $499/month running 24/7, that works out to $0.69/hour. The federal minimum wage is still $7.25/hour, ten times more.

What changed is that it’s no longer one company offering one robot. The market has stratified into tiers, prices are compressing, and Chinese manufacturers are shipping units at volumes that would have seemed absurd 18 months ago.

Unitree’s G1: Available for approximately $16,000, cheaper than a used Honda Civic. AgiBot shipped over 5,100 units in 2025 alone. UBTECH added another 1,000+. These aren’t pre-orders or prototypes. They’re invoiced, delivered, operational robots.

When a humanoid robot costs less than six months of minimum-wage labor, the economic logic becomes unanswerable. Not in theory, but in spreadsheets, in procurement decisions being made right now.

The price curve underneath all of this is steep: the average humanoid robot ran about $85,000 in 2023 and roughly $25,000 in 2025, a 70% drop in two years as the market split into tiers. On that trajectory a basic humanoid soon costs what a dishwasher does today, and nobody flexes about owning a dishwasher.

The Big Players: 2026 Status Report

Physical Intelligence: The Robot Brain Factory

What they do: Foundation models for robots - “ChatGPT, but for robots,” as co-founder Sergey Levine puts it.

The money: In March 2026, Physical Intelligence entered talks to raise $1 billion at an $11 billion valuation, doubling their $5.6 billion valuation from just four months earlier. Founded in 2024 by former Google DeepMind researchers and Stanford/Berkeley academics, they’ve raised $1.1 billion to date. Founders Fund, Lightspeed, Thrive Capital, and Lux Capital are all in.

Why it matters: Physical Intelligence is building the mind that goes into the robot, not the robot itself. Their approach mirrors what OpenAI did for language: create a general-purpose foundation model that any hardware manufacturer can deploy. If they succeed, every humanoid from every manufacturer gets smarter simultaneously. This is the “Android moment” for robotics, the platform layer that commoditizes hardware and concentrates value in software.

Physical Intelligence is one of 27 “Physical AI” startups that quietly raised $50M+ in Q1 2026 alone. The investment thesis is clear: the robotics bottleneck has shifted from hardware to intelligence. Solve the brain, and every body benefits.

Google DeepMind: The Kingmaker Strategy

Rather than building humanoid robots, Google is doing something potentially more powerful: becoming the intelligence layer for everyone else’s robots.

In January 2026, Boston Dynamics and Google DeepMind announced a partnership to integrate Gemini Robotics foundation models into the new Atlas robots, unveiled at CES 2026 in production-ready form. All Atlas deployments for 2026 are already fully committed, shipping to Hyundai’s Robotics Metaplant and Google DeepMind facilities.

Then in March 2026, Agile Robots, with 20,000+ deployed systems globally, partnered with Google DeepMind to deploy Gemini Robotics models across electronics manufacturing, automotive, data centers, and logistics.

The pattern: Google is running the Android playbook for robotics. Build the OS, not the phone. Let Boston Dynamics, Agile Robots, and whoever comes next compete on hardware while Google captures the intelligence layer. The data from every partnered robot feeds back into Gemini, making the models better, which attracts more partners, which generates more data.

If this sounds familiar, it’s because this is exactly how Google won mobile. And the robotics market is projected to be far larger.

Gemini Robotics 2 (July 30, 2026): DeepMind shipped the next version, and the headline feature is whole-body control — one policy coordinating legs, torso, and arms together, instead of a mobile base carrying an independently-controlled pair of arms. Launch partners are Apptronik, Boston Dynamics, and Agile Robots; the demos run on Apptronik’s Apollo 2.

The genuinely useful thing DeepMind did was publish the success rates instead of the highlight reel. On Apollo 2, whole-body picking scored 76.3% from a shelf, 68.4% from a table, and 45.7% from the floor. Parallel-gripper work did better: 74.2% on general pick-and-place, 78.9% on tool kitting, 89.6% on precise insertion. Multi-finger dexterity swings wildly by task, from 32% to 92% — unscrewing a light bulb hit 92%, a task needing coordinated full-body positioning.

Those numbers are the most honest thing published about humanoids all year, and they reframe the whole timeline debate. A robot that picks correctly 45.7% of the time off the floor is not a worker. It is a worker plus a supervisor, and the supervisor costs more than the robot saves. Every occupation has a per-task success rate below which automation adds labor. Publishing the rates turns “robots are coming for your job” from a vibe into a countdown you can actually check.

Figure AI: From Factory Floor to White House

The achievement: Figure AI went from an 11-month BMW factory pilot in 2025 to a $39 billion valuation and a White House visit in under a year.

In October 2025, Figure unveiled Figure 03, a complete hardware and software redesign featuring Helix, their proprietary vision-language-action AI. The specs represent a generational leap:

Double the frame rate, one-quarter the latency, 60% wider field of view
Palm cameras and tactile sensors detecting forces as small as three grams (the weight of a paperclip)
Soft textile coverings, wireless charging - designed for homes, not just factories
Target price: $20,000

In September 2025, Figure raised more than $1 billion in Series C from NVIDIA, Intel Capital, Qualcomm Ventures, and Salesforce. Post-money valuation: $39 billion, a 15x increase from the $2.6 billion Series B just 18 months earlier.

And then, on March 26, 2026, Figure 03 appeared at the White House alongside First Lady Melania Trump during the Fostering the Future Together Global Coalition Summit. A humanoid robot as a guest at the White House. The symbolism writes itself.

What Helix changes: Previous robots required explicit programming or teleoperation. Helix is a generalist vision-language-action model that can observe its surroundings, respond to natural language, and interact with the physical world without extended training. You don’t program Figure 03. You talk to it. The barrier to automation collapsed from “hire a robotics engineer” to “hire anyone who speaks English.”

The BotQ ramp (May 2026): Figure backed the valuation with throughput. Its BotQ factory went from one robot per day to one robot per hour in under four months. That 24x jump put 350+ Figure 03 units on the floor by late May, with CEO Brett Adcock saying the line was on pace for 55 robots that week and end-of-line yields already above 80%. BotQ was built for 12,000 units a year, and the point is to harvest data rather than to undercut Unitree on price. Every robot shipped is a training run for Helix; each doubling of output makes the next one smarter and cheaper to build. The companies still hand-assembling prototypes are watching their window close.

Tesla Optimus: The Scale Machine Awakens

The 2025 reality check we wrote: Musk overpromised. A few hundred units. Gen 3 delayed.

The 2026 reality (so far): The headline that ran in January, “Tesla commences Optimus Gen 3 mass production at Fremont,” turned out to describe the hands, not the robot. What entered production on January 21, 2026 was the upgraded Gen 3 hand system (22 degrees of freedom, 50 actuators each); the body it bolts onto is still the Gen 2 design. On the Q4 2025 earnings call (January 28, 2026), Musk conceded that the few hundred Optimus units built so far are “primarily for learning and data collection rather than performing productive tasks.” In plainer terms, none are doing useful work on the factory floor yet.

The full V3 body redesign, lighter and faster and cheaper to build, keeps slipping. In April, the reveal was scheduled for late July or August 2026, with robot production starting at Fremont the same summer, on the line freed up when the last Model S and X rolled off in early May. Musk declined to give any 2026 production target, calling the rate “literally impossible to predict” for a machine with 10,000 unique parts on a brand-new line, and said the first units would learn “simple skills in the factory” before scaling.

Then the window closed with nothing in it. July came and went without a V3 reveal. Tesla’s Q2 shareholder update on July 22 confirmed the Model S/X lines at Fremont are decommissioned and the first Optimus lines are being installed, with production “anticipated later this year” — the third time in eighteen months the date has moved right. Musk said on July 1 that production had not begun and that initial output would be extremely slow. A second Optimus factory at Giga Texas is under construction, with production expected around summer 2027.

The numbers that matter (revised):

2026 production target: none given. Musk stopped quoting a number after missing his January 2025 “10,000 robots this year” promise by all 10,000
V3 reveal: missed the July/August window; as of August 1, 2026 the Fremont line is still being installed and production is “anticipated later this year”
Long-term ambition (firmly in Musk-time): Fremont at 1 million units/year, a Giga Texas plant designed for 10 million units annually, target cost $20,000/unit, consumer sales around 2027

Hold that against the Chinese numbers below. In the same eighteen months Tesla spent moving one reveal date, Unitree shipped its eleven-thousandth humanoid and filed to go public.

The pattern is the Tesla playbook: announce aggressive timelines, miss them by 12-24 months, but eventually ship at scale. The first million miles of Autopilot data were terrible. The billionth mile created the world’s most valuable driving AI. The same fleet-learning flywheel is now spinning for physical labor, just slower than promised.

Tesla has posted 100+ job openings for Optimus production, from AI engineers to manufacturing specialists. They’re not prototyping anymore. They’re industrializing.

Boston Dynamics Atlas: The Google-Powered Athlete

Boston Dynamics, now owned by Hyundai, unveiled the production-ready Atlas at CES 2026. The new Atlas is fully electric, lifts 110 pounds, and is powered by Google DeepMind’s Gemini Robotics models.

All 2026 Atlas deployments are already committed, primarily to Hyundai’s automotive manufacturing and Google DeepMind’s research facilities. The partnership is expected to become “a driving force of manufacturing transformation, beginning in the automotive industry.”

Atlas has always been the most physically capable humanoid robot in existence. The difference now is that it has a brain to match, plus a manufacturing giant (Hyundai) behind it with the scale to actually produce them.

1X Neo: The Consumer Pioneer

1X opened pre-orders in late October 2025 for the world’s first consumer-ready home humanoid, with confirmed 2026 delivery. The pitch remains compelling: 66 pounds, lifts 154 pounds, quieter than your refrigerator (22 dB), $499/month.

The “human-in-the-loop” strategy, where remote operators train the AI through real-world task execution, is proving prescient. Rather than waiting for full autonomy (which no one has achieved for home environments), 1X ships a usable product now that gets smarter every day.

The EQT partnership to deploy 10,000 units across 300+ portfolio companies by 2030 remains the largest commercial commitment for consumer-grade humanoids.

Amazon: Acquire and Integrate

The companies above build robots or the brains for them. Amazon is taking a third route: buy the startups and absorb them into the largest automated logistics network on Earth. On March 24, 2026, CNBC reported Amazon acquired Fauna Robotics, the two-year-old New York startup behind “Sprout,” a 50-pound, 3.5-foot bipedal robot built for tight spaces. It was Amazon’s second robotics acquisition that month.

Sprout’s value is its geometry. Amazon already runs hundreds of thousands of wheeled, Kiva-descended pods (it bought Kiva Systems in 2012 for $775 million) that carry shelving across a flat floor but can’t climb, reach, or pick. A bipedal unit handles elevation changes and the chaos between multi-level shelves. The wheeled robots solved the warehouse floor; Sprout is aimed at the shelves and the “last ten feet.”

This is the AWS playbook applied to bodies: commit early, absorb the engineering teams, integrate into a machine that already spans the globe. While Tesla builds Optimus from silicon up and 1X sells to consumers at $499/month, Amazon needs only a robot that can pick, pack, and sort inside its own fulfillment centers. The strategic tell is the timing: Amazon cut 14,000 corporate jobs in 2025, its largest single layoff of the year, while pouring money into AI and robotics. The company employs roughly 1.5 million people and has been the rare large employer that kept hiring humans as it added robots. If its bipedal push changes that, the labor cliff loses its biggest counterexample.

The Chinese Juggernaut

The headline that should alarm Western policymakers: Chinese companies control approximately 90% of the global humanoid robot market.

In 2025, the top three humanoid robot shippers were all Chinese:

Company	2025 Shipments	Market Share
AgiBot (Shanghai)	5,100+	~39%
Unitree (Hangzhou)	4,200-5,500	~33%
UBTECH (Shenzhen)	1,000+	~8%

China’s humanoid sector now counts over 160 companies. Analysts have stratified the market into three tiers, with the weakest players already being forced out in 2026.

The cadence of that output is the part worth watching. AgiBot rolled its 10,000th unit off the line on March 30, doubling its cumulative output in a single quarter — then hit 15,000 in late June, less than three months later. Meanwhile Beijing began drafting industry-wide engineering standards for humanoid robots: reference architectures, joint design specifications, communication protocols, certification requirements. You standardize an industry when you’re confident you’re leading it. If your robot doesn’t conform to these specs, it doesn’t get certified for the Chinese market, and increasingly, for markets that adopt Chinese standards abroad. It’s the Huawei 5G playbook applied to bipedal machines — and as of July 28, Washington is running the same play in the opposite direction.

The buyers are institutional now, not experimental. In the first half of 2026, State Grid Corporation — China’s national electricity utility — placed a single procurement order worth RMB 6.8 billion covering 500 humanoids, 3,000 dual-arm robots, and 5,000 quadrupeds. That is a utility buying a robot workforce the way it buys transformers.

Unitree goes public (July 2026): CEO Wang Xingxing confirmed a 20,000-unit shipment target for 2026, roughly four times the 2025 output, and by early June the company had built about 11,000 bipedal humanoids cumulatively. Then it did the thing no humanoid maker had done: it launched its IPO on Shanghai’s STAR Market on July 31, selling 40.45 million shares (10% of enlarged capital) to raise RMB 4.2 billion at a RMB 42 billion valuation — about $620 million at roughly $5.9 billion. Regulatory approval came 104 days after filing, a STAR Market record.

The prospectus is the real story. Humanoids are now ~52% of Unitree’s revenue, which reached RMB 1.7 billion on a 227% compound growth rate, with core net profit of RMB 591 million at 60% gross margins. Sixty percent. A hardware company selling bipedal robots is running software-company margins while the American entrants are still pre-revenue. With the G1 priced at ~$13,500 (revised down from $16K) and shipping at scale, Unitree is about to set the first public-market valuation benchmark for the entire category.

This is the EV playbook all over again. China identified humanoid robots as a strategic industry, marshaled state support and manufacturing scale, and is now running the same strategy that gave it 60%+ of global EV production. By the time Western competitors reach scale, Chinese firms will have years of manufacturing learning and deployment data.

Europe’s counter-move (May 2026): The first concrete sign the West won’t cede the whole field came from Germany. Bosch, the world’s largest auto-parts supplier, agreed to manufacture humanoid robots for the UK startup Humanoid, after its HMND 01 units autonomously moved boxes through a Bosch intralogistics facility in Bühl. Schaeffler signed on as preferred actuator supplier, covering over half of Humanoid’s joint demand through 2031, a seven-digit number of actuators, with first deployments running December 2026 through June 2027 at two German plants. A two-year-old startup just acquired two Tier-1 industrial partners and a production timeline. It won’t out-scale Shenzhen overnight, but when Bosch allocates manufacturing capacity to a product category, its entire customer network notices.

Washington’s Answer: The Certification Chokepoint

On July 28, 2026, the FCC added foreign-produced advanced robotic devices and connected power inverters to its Covered List, after national security determinations found both categories pose “unacceptable risks to the national security of the United States or the safety and security of United States persons.”

The definition is broad on purpose. An “advanced robotic device” is any mechanical mobile device — humanoids and quadrupeds explicitly named — that can locomote, avoid obstacles, or navigate; that operates at a distance from a human operator on commands or sensor data; and that weighs more than 4.4 pounds. Covered models generally cannot obtain the FCC equipment authorization required before a device may be imported, advertised, or sold in the United States. The rule bites only on new models seeking authorization; robots Americans already own and models previously approved are untouched. A foreign-produced robot can still get conditional approval after a favorable national security determination from the Department of War.

Read that against the market share table above. Chinese firms hold roughly 90% of global humanoid shipments and control the low end of the price curve. The United States did not tariff them, out-manufacture them, or beat them on cost. It simply revoked their permission slip.

This is the move the Unscarcity framework keeps predicting and the one most people miss. When a scarce input gets manufactured rather than mined, control migrates away from the resource and toward the right to authorize a device for a market. There is no humanoid equivalent of an oil field to seize. There is a certification regime, and whoever administers it decides which bodies are allowed to do work inside a border.

Two consequences follow. The first is that the American humanoid industry just received the most valuable subsidy available, which is not a tax credit but the removal of its competition. That buys Figure, Apptronik, Tesla, and Boston Dynamics time they were visibly losing on the manufacturing curve. The second is that abundance itself gets partitioned. If the machines that make labor cheap can only cross borders their governments approve, then the arrival of post-scarcity stops being a technology date and becomes a trade-policy date, different in every jurisdiction. The robots are ready before the politics is, which is the same sentence this project keeps writing about every other abundance technology.

Note which category rode along in the same order: connected power inverters, the boxes that tie solar panels to the grid. Bodies and energy, regulated in a single proceeding. The two physical bottlenecks on abundance, treated by the state as one problem — because they are.

Why 2026 Is Different From 2025

In 2025, we wrote that three things had converged: AI brains, battery muscle, and scaled manufacturing. That was true but incomplete.

What changed in 2026 is the fourth convergence: the business model crystallized.

Foundation models for robots (Physical Intelligence, Google DeepMind Gemini Robotics) mean hardware manufacturers don’t need to solve AI; they can license it. This lowers the barrier to entry dramatically and accelerates iteration.
The talent pipeline opened. Physical Intelligence alone drew from DeepMind, Stanford, and Berkeley. Tesla posted 100+ Optimus jobs. Every major robotics company is hiring at scale. The field went from “research curiosity” to “career path” in 18 months.
The customers showed up with purchase orders, not press releases. Google’s 200 MW fusion PPA for CFS. Microsoft’s 50 MW Helion deal. Hyundai’s full Atlas commitment. EQT’s 10,000-unit Neo deal. BMW’s Figure deployment. These are legally binding contracts with penalties for non-delivery.
China forced the pace. With 90% market share and aggressive pricing ($16,000 Unitree G1), Chinese manufacturers made “wait and see” a losing strategy for every other manufacturer on Earth.

The Bottleneck Isn’t Bodies. It’s Data.

The mechanical bodies are ready and the foundation models exist. What’s missing is the demonstration data that connects them, and that is the real pacing constraint on the Labor Cliff.

Language AI had the internet: trillions of words humans had already written, lying around for the scraping. There is no equivalent corpus for physical tasks. Nobody uploaded the exact finger pressure to crack an egg one-handed or the wrist rotation to fold a fitted sheet. Physical skill has always passed through apprenticeship, not publication, so it lives untranscribed in human muscle memory.

Tesla’s self-driving program is the trick everyone is trying to repeat. Cars shipped from 2016 with a second GPU that did nothing but watch the driver. By 2024 the fleet had logged over 35 billion miles of real-world driving data, with customers paying $35,000-$100,000 for a car they wanted and generating Tesla’s most valuable asset as a byproduct. The collection was passive, the task was narrow, and by 2025 over 7 million vehicles were doing it. None of that transfers cleanly to a home robot. You can’t passively collect manipulation data, so 1X’s “human-in-the-loop” approach pays operators to teleoperate robots, making cost scale up with data volume. The task space is thousands of chores, not one structured activity. Even the bullish 50,000-shipment projection for 2026 is a 140x gap behind Tesla’s 7 million cars.

One attempt removes the robot from the training loop entirely. On April 2, 2026, Forbes reported on a startup called Generalist building sensor-packed gloves that capture sub-millimeter finger positioning, grip force, and release timing as a human does the task, then translate it into robot training data. A hundred warehouse workers in gloves for a month could in principle out-produce a fleet of teleoperated robots over a year. The open problem is transfer: human hands and robot grippers have different kinematics, so captured human motion is worthless if mapping it to a robot’s actuators adds too much noise.

This bottleneck is, perversely, good news for the book’s argument. It buys a window, maybe five years, maybe ten, to build the institutional infrastructure before robots replace physical labor at scale. The window is closing: when one of these data approaches succeeds, and one will, the floodgates open.

Why This Changes the Equation

1. The Labor Cliff Is Physical

The “Labor Cliff” discussion usually fixates on white-collar knowledge workers losing ground to AI. But there are 3.5 billion employed humans on Earth. The majority work in informal sectors - agriculture, construction, domestic work - where AI chatbots are irrelevant.

These workers aren’t threatened by language models. They’re threatened by $16,000 bipedal robots.

When economists talk about “technological unemployment,” they usually imagine coders and accountants. The humanoid robot makes the threat concrete for the factory worker, the warehouse picker, the home aide, the janitor - the “bottom rungs” that have traditionally absorbed displaced workers.

If those rungs disappear, there is no “reskilling” path that moves 2 billion people into knowledge work. The only coherent response is decoupling survival from labor entirely. Welcome to the Abundant Foundation.

2. Solving the Demographic Collapse

Japan’s demographic crisis has only deepened. Over 10% of Japanese are now 80+, with a projected 11 million worker shortfall by 2040. China is racing toward “hyper-aged society” status (over 20% aged 65+) by 2030.

The math doesn’t work. You cannot care for aging populations with shrinking workforces using human labor alone. Humanoid robots provide labor “liquidity” that biology can no longer supply. One robot per elderly household is the difference between dignity and neglect when there simply aren’t enough young people to go around.

This explains why Asia Pacific leads humanoid adoption: demographic necessity, not technological enthusiasm.

3. The End of “Dangerous, Dirty, Dull”

The World Health Organization reports that 2.4 billion workers suffer productivity losses from extreme heat, with performance dropping 2-3% for every degree above 20°C. In a warming world, this isn’t improving.

In an Unscarcity civilization, no human has to enter a coal mine, process meat in a slaughterhouse, or stand on an assembly line for 12 hours. The shift runs from Operator to Architect, not from “worker” to “unemployed.” You stop being the paintbrush and start being the artist.

The Market Speaks (Loudly)

Metric	Figure	Source
2025 global shipments	~16,000 units	TrendForce
2026 projected shipments (global)	50,000+ units	TrendForce
2026 projected shipments (China alone)	50,000 units	Morgan Stanley (June 2026, doubled from 28,000)
2026 market size	$4-6 billion	Multiple analysts
China market size, 2030	$15 billion / 446,000 units/yr	Morgan Stanley
2027 projected shipments	115,000 units	ABI Research
2030 projected shipments	195,000 units/year	ABI Research
2035 projected market	$38 billion	Goldman Sachs
2050 projected market	$5 trillion	Morgan Stanley
Projected units by 2050	~1 billion	Morgan Stanley

One billion humanoid robots, for a global population of ~10 billion. Call it what it is: a parallel labor force, larger than any single nation’s workforce, operating 24/7 without wages, benefits, or retirement.

Healthcare (elder care, rehabilitation, surgery assistance), logistics, manufacturing, and domestic service are expected to be the dominant sectors. The geography of this transition will reshape geopolitics as profoundly as the oil economy did in the 20th century.

The Question We’re Still Not Ready For

The arrival of the $16,000-$20,000 humanoid robot, available today and improving monthly and shipping at scale, is the final nail in the coffin of the “labor-for-survival” social contract.

We built our entire civilization on the assumption that human labor was the scarce input into the economy: our concept of “earning a living,” our moral framework around “work ethic,” our social hierarchies of “productive” vs. “unproductive” citizens.

That assumption dissolved in 2025. In 2026, we’re watching the aftermath, and watching governments discover that the assumption’s replacement is a customs question.

50,000 units from China alone this year, 195,000 globally by decade’s end, a billion by 2050. The question is no longer “how do we get the work done?” It is “how do we organize society when the work is already done?” — with a newer, uglier question stacked on top: whose society gets the machines first, and who decides.

The answers - the Abundant Foundation, Impact, Civic Service, the MOSAIC governance structure - are what the rest of this project explores. The robots have shipped. The countdown is well underway.

References & Further Reading

Internal Links:

External Sources:

This article was last updated August 1, 2026 - added the FCC’s July 28 Covered List order blocking new foreign humanoid and quadruped models from US equipment authorization, DeepMind’s Gemini Robotics 2 with its published per-task success rates, Morgan Stanley’s second doubling of its China forecast to 50,000 units, Unitree’s July 31 STAR Market IPO launch and prospectus economics, AgiBot’s 15,000th unit, and Tesla missing the July/August Optimus V3 window. Previous version covered 2025 data as “humanoid-robots-2025.”

The robots have shipped. Are you ready?

Humanoid Robots 2026: 50,000 Units and a US Import Ban