What this chapter covers
This chapter, “Humanoid robots at home and at war,” covers where human-shaped robots stand: what is deployed for pay today, how current AI models change what a robot body can do, and the labor, safety, and control questions that follow once such a machine works in a warehouse, factory, or home. Later pages go deep on domestic deployment, privacy and control, military use, and labor economics. This page opens with the plainest question: separate from demo reels and funding announcements, what is a humanoid robot actually doing today, and who is paying for it?
This page distinguishes company self-reporting, analyst forecasts, and independently reported tests. Figures described as a company's own claim have not been independently audited unless stated otherwise. Every dated figure reflects public reporting available as of September 13, 2026.
The strongest real case: a warehouse robot on a payroll
The most credible “actually working, actually paid” deployment belongs to Agility Robotics’ Digit, not to any of the more famous consumer-facing names. GXO Logistics signed what it called the industry’s first multi-year robots-as-a-service agreement for a humanoid robot, and Digit began commercial operations at a GXO facility near Atlanta on June 5, 2024, later expanding to Spanx’s Flowery Branch, Georgia facility (GXO announcement). Agility’s own count, published in November 2025, put the total at more than 100,000 totes moved at that site (Agility Robotics). By 2026 the company describes Digit as deployed across nine customer facilities — including GXO, Schaeffler, Toyota Motor Manufacturing Canada, and Mercado Libre — with more than 65,000 cumulative operating hours. Those totals are Agility’s own reporting, not an independent audit.
The same company’s finances complicate the success story. Agility is going public by merging with a special-purpose acquisition company, Churchill Capital Corp XI, rather than through a traditional IPO, at a reported $2.5 billion pre-money valuation. Its own 2026 SEC filing disclosed a $138.1 million loss for 2025 and stated plainly that, without the merger, “there is substantial doubt about its ability to continue as a going concern” (SEC EDGAR filing). Real commercial deployment has not yet translated into a comfortable balance sheet, even for the field’s strongest case.
The consumer-hype layer: Tesla’s own contradiction
Tesla’s Optimus is the most-hyped humanoid robot and, by the company’s own account, the least commercially real. In January 2025, Elon Musk said Tesla would have “several thousand Optimus robots doing useful things” by year’s end. On Tesla’s Q4 2025 earnings call, he reversed that: “We are still very much at the early stages of Optimus. It’s still in the R&D phase… It’s not in usage in our factories in a material way” (Electrek).
The gap between demo and reality showed up on stage, too. At a Tesla “Autonomy Visualized” event in Miami in December 2025, an Optimus unit fell backward while handing out water bottles, its hand motions resembling removal of a VR headset — fueling speculation a hidden operator was driving it. Electrek called the moment a “Wizard of Oz reveal” (Electrek; Fortune). The next hardware generation, Optimus V3, has slipped from a Q1 2026 reveal to “the middle of this year,” with Musk declining to give a 2026 volume target, calling it “literally impossible to predict”; production is now aimed at Fremont around late July or August 2026 (Electrek).
Figure AI: a real factory pilot, unaudited numbers
Figure AI has the field’s most concrete industrial pilot outside Agility’s. Figure 03, running the company’s “Helix 02” vision-language-action model, arrived at BMW’s Plant Spartanburg on June 30, 2026 for a sequencing task — sorting components into trolleys (Figure AI; BMW Group). That builds on Figure 02, which Figure’s own release says “contributed to the assembly of 30,000 cars” in 2025 — a company claim with no independent performance or error-rate data behind it. Investors priced the story generously regardless: a September 2025 round led by Parkway Venture Capital valued Figure at $39 billion, up from $2.6 billion eighteen months earlier (TechCrunch).
1X NEO: a home robot you can order today
1X Technologies is the one company selling a humanoid robot for the home, and NEO is the subject of this chapter’s next page in full. In short: NEO costs $20,000 outright or $499 a month, both requiring “Expert Mode” — human teleoperation — with first U.S. deliveries expected in the third or fourth quarter of 2026 (1X). When the Wall Street Journal’s Joanna Stern tested it, every household task NEO performed was driven by a remote human operator; retrieving a water bottle took one minute and seven seconds. The next page, domestic robots: capability and reality, covers NEO’s teleoperation model and privacy mitigations in full.
The industrial pack: Boston Dynamics, Apptronik, and China
Boston Dynamics unveiled a production, fully electric Atlas — 56 degrees of freedom, 50-kilogram payload — at CES on January 5, 2026, without disclosing a price. Every 2026 Atlas unit is already committed to Hyundai’s Robotics Metaplant Application Center and Google DeepMind, with Hyundai targeting roughly 30,000 units a year by 2028 (Boston Dynamics). Apptronik’s Apollo has a real, narrow pilot with Mercedes-Benz — parts delivery, not assembly, at two European plants — with commercial-quantity deployment targeted for 2027; Apptronik closed a $520 million round at a $5 billion valuation in February 2026 (CNBC).
China dominates by volume. Unitree’s G1 ($16,000, launched May 2024) and fellow maker AgiBot together took 81 percent of global humanoid shipments in 2025, per Omdia data reported by Forbes, with China holding roughly 90 percent of the global market (Forbes). Xpeng’s “Iron” drew public disbelief that a robot, not a person in a suit, walked on stage at its November 2025 unveiling — founder He Xiaopeng cut it open on camera to prove it wasn’t (CNN Business). UBTech’s Walker S2 has begun mass production, targeting 500 units in 2025–2026 with orders reportedly exceeding 800 million yuan (PR Newswire).
What analysts actually project
Goldman Sachs raised its 2026 baseline shipment forecast to 75,000 units and its long-range forecast to 890,000 by 2030 and 6.5 million by 2035, a roughly $138 billion market. Morgan Stanley separately forecasts China’s humanoid market reaching $15 billion by 2030 — 446,000 units, a 106 percent compound annual growth rate — and the global market reaching $5 trillion by 2050. Both banks hedge the home-adoption question explicitly. Morgan Stanley’s analysts state it plainly: “We are not going to see a robot in every home overnight,” adding that general-purpose home capability needs “about another decade” (CNBC).
The credentialed skeptics
Rodney Brooks — MIT professor emeritus, co-founder of iRobot and Rethink Robotics — calls the humanoid investment climate a bubble in waiting, singling out Tesla’s and Figure’s approach of teaching dexterity from human video as “pure fantasy thinking,” since human hands carry roughly 17,000 specialized touch receptors no robot hand replicates (TechCrunch). Asked about human-shaped machines matching movie robots, he told the Boston Globe, “Check back in 15 years,” predicting today’s bipedal form factor may itself be abandoned for wheels (Boston Globe).
IEEE Spectrum’s Evan Ackerman makes a related point: the gap between edited demo footage and real-world performance is large, and battery life, industrial uptime expectations, and immature safety standards are concrete blockers independent of the AI’s capability. Former Agility chief product officer Melonee Wise put it more bluntly: “The bigger problem is demand — I don’t think anyone has found an application for humanoids that would require several thousand robots per facility” (IEEE Spectrum).
Safety incidents, reported thinly so far
Two 2026 incidents are worth naming honestly rather than omitting for lack of a tidy source. At the World Robot Conference in Beijing on August 19, 2026, multiple humanoid robots malfunctioned on stage; one lost balance after its remote-control signal failed, raising questions about safety power-cutoffs. No top-tier wire-service account has surfaced; the fullest report available is regional press (The News International). Separately, the OECD’s AI incidents database logged a February 5, 2026 case in which a humanoid robot demonstration in China collided with and injured an elderly bystander — a structured database entry rather than a full news account (OECD.AI). Neither should be read as more thoroughly documented than it is.
The bottom line
Real, paid, commercial humanoid deployment exists, and it is overwhelmingly industrial and warehouse work, not home use — Agility’s GXO contract is the strongest case, and even it carries a going-concern warning. The most-hyped consumer-facing players either admit they remain in an R&D phase, like Tesla, or depend on human teleoperation behind the curtain, like 1X. Credentialed skeptics such as Rodney Brooks push mainstream robot dexterity out past a decade, and no major bank or roboticist currently projects mass-market home humanoid adoption before the early-to-mid 2030s. The rest of this chapter builds on that baseline, starting with the home, where the gap between marketing and reality is widest.