AI Is Getting a Body

Since early 2025, a humanoid robot named Digit has been working eight-hour shifts at a Schaeffler bearing plant in Cheraw, South Carolina. At a GXO warehouse in Flowery Branch, Georgia, the same robot has moved more than 100,000 totes. Digit does not appear on either company’s equipment ledger, because neither company bought it. Instead, Agility Robotics rents the robot out through Robots-as-a-Service contracts and bills a recurring rate, the way a staffing agency bills for a temp.

The economics behind that arrangement are, in our view, the most interesting numbers in robotics today, provided you keep two of them separate. The first is what it costs Agility to run the robot: $10 to $12 an hour today by the company’s own account, with a projected path down to $2 or $3 as production scales. The second is what the customer pays: a subscription that Agility’s investor materials model at roughly $8,500 a month per robot, benchmarked against a fully loaded human labor cost of about $30 an hour. Agility’s margin lives in the gap between those two numbers, and the customer’s case lives in the gap between the subscription and the wage.

Two qualifications before we build anything on this. First, Agility is heading to Nasdaq through a roughly $2.5 billion SPAC, so these are a seller’s published numbers on the way to a listing, and we discount them accordingly. Second, an hourly figure means little until it is adjusted for the work actually completed: a robot working at half a human’s pace effectively costs twice its stated wage. The honest metric, and the one we will keep returning to, is cost per completed unit of work, not cost per hour. What we don’t discount is the structure of the deal. For the first time, a robot is priced, invoiced, and renewed like an employee.

This is happening in an explosive year when everything else in robotics is getting louder, too. Venture funding into the category broke the full-year record by mid-2026. The firm that told us software was eating the world (a16z) raised a $1.1 billion fund devoted to the physical buildout of AI, robotics included. A humanoid maker went public in Shanghai and closed its first day up 460%. And Amazon quietly passed one million deployed warehouse robots, a milestone that turned deployment from a prediction into a baseline.

A year like this often produces more headlines than understanding. So rather than narrate the news, we have distilled it into the five ideas we most want investors to walk away with. These observations come from our own experience investing in the category, multiple conversations with folks in our expert network, and conversations with fellow robotics-focused investors.

1. Robots Are Priced Against Wages Now, Not Equipment

For thirty years, the standing joke was that robotics is where venture capital goes to die, and for thirty years it was more or less earned. A robot was a machine tool. It was sold as capital equipment, sized against an equipment budget, and paid back over years, and every deployment was a bespoke engineering project. A market defined that way is structurally small, and no amount of enthusiasm will make it bigger.

What changed is how the robot gets paid for. Digit has no sticker price in any meaningful sense. It has a monthly rate and an operating cost, benchmarked against a human wage, with a stated payback of under two years against a $30-an-hour worker, per Agility’s investor materials. These are not science projects, either. The GXO and Schaeffler deployments we opened with are commercial contracts measured in totes moved and shifts covered, and Toyota’s Canadian plant added a third this February.

The buyer has changed just as much as the pricing. Deloitte and The Manufacturing Institute project that U.S. manufacturing may need 3.8 million new workers by 2033, and that nearly 1.9 million of those roles could go unfilled. Almost two thirds of manufacturers already call talent their top business challenge. For that buyer, the robot is not competing with a wage. It is competing with an empty shift and zero output, and zero output loses every negotiation.

This is why we size the market against payroll rather than equipment, a thesis we call Payroll-TAM. The addressable market is the share of the wage bill attached to dull, dirty, dangerous, and unfillable work, and it grows as the robot’s effective wage falls. At $12 an hour of completed work, robots win totes and food lines. At $6, they win most of the warehouse. At $3, they start winning everything adjacent. Every robotics forecast built on equipment sales is not just low, but it’s measuring the wrong thing entirely.

Software ate margins. Robots eat payroll.

2. Narrow Applications Are Winning. General Purpose Isn’t There Yet.

Look closely at where robots actually earn money today, and a pattern emerges: every one of them has a job description. Amazon’s million-robot fleet is an armada of specialists. Digit’s paid hours come almost entirely from one task, moving totes. The robots generating revenue in 2026 do one thing, in one environment, extremely well.

The companies that promised everything have had a harder year. K-Scale Labs shut down in November 2025, and Cartwheel Robotics followed in February. Sanctuary AI gave up on its hardware in June and pivoted to software. Amazon quietly absorbed Fauna Robotics. None of this means the market is rejecting robotics. It means the market is sorting demonstration from deployment, and the sorting favors focus.

Our portfolio is built around that discipline. We back companies where the wage math closes on a specific task today and treat general-purpose ambition as upside rather than the reason to invest. Chef Robotics automates high-mix food manufacturing without requiring customers to retrofit their lines. Dextrous Robotics unloads boxes from shipping containers, one of the highest-turnover jobs in the supply chain. Dyna Robotics pairs foundation models with low-cost hardware across foodservice, grocery, and hospitality, and PSYONIC already sells the first commercially available bionic hand with lifelike touch feedback. The same logic extends to defense, where the customer buys autonomy for a defined mission and treats it as a national-security priority rather than a cost savings. Defense was PitchBook’s largest robotics segment last year at $8 billion, and it is where our position in Ghost Robotics, the legged-vehicle maker backed alongside In-Q-Tel, exited in 2024. Nobody dreams of unloading a shipping container. That is precisely the point.

To be clear about what this discipline is and is not: it is not a bet against general-purpose robots. It is a refusal to pay today’s prices for tomorrow’s capabilities. Where robots earn wages now, we underwrite the wage math. Where they might someday, we underwrite the entry price. That distinction is the whole difference between idea two and idea five, and we will come back to it.

3. The Robot Body Is Commoditizing. There Are Still Returns in Software and Data.

The clearest lesson of the past twelve months came from Hangzhou. Unitree shipped more than 5,500 humanoids in 2025, sells its flagship G1 model from about $13,500, and still reported first-quarter adjusted profit down by roughly half as competition intensified and R&D spending surged. Even the most celebrated name in the category is feeling the same gravity: Figure raised at $39 billion last September, and by this summer its shares were trading below that mark on the secondary market. The robot body is becoming a commodity, and it is happening quickly.

At first glance that seems to threaten our thesis. If bodies get cheap, why wouldn’t the robot’s wage get cheap too? In time it will, and that is actually the engine of the whole opportunity, because a falling wage is what expands the market. The question for investors is simply who keeps their margin while it happens. Our answer is whoever owns the things a cheap body cannot provide. A $13,500 body is a commodity. A robot that has run your food line for eighteen months, learned your products, cleared your safety audit, and comes wrapped in a service network your operations team trusts is not.

The deepest of those moats is data. Language models had the internet to scrape, but there is no internet of physical skills. Robot intelligence is trained on some blend of human video, simulation, demonstrations, teleoperation, and data flowing back from robots already in the field, and the market research we reviewed this summer, drawing on dozens of practitioner interviews, found that no single source has won. The most valuable of them, the data from deployed fleets, is never for sale at all. It belongs to whoever operates the robots, which means every shift a deployed robot works deepens a moat its competitors cannot buy.

For a portfolio, this argues for owning the layers that get paid no matter whose body wins. RGo Robotics supplies full-stack visual perception and is backed alongside us by Qualcomm Ventures and Nvidia. Vayu Robotics delivers computational imaging that outperforms LiDAR at one-fiftieth the cost, with Khosla Ventures on the cap table. The same falling sensor prices that squeeze the body-makers are, for these companies, a product roadmap.

4. The Demo Is Not the Product. The Deployment Is.

Every robotics company can make one robot perform one task in one building with three engineers watching. The hard part, and the part that separates companies from science projects, is making a robot work every day, at a customer, at a profit. A company only becomes real when integration, uptime, and field service hold up across many sites, because that is where the unit economics survive contact with the customer, or don’t.

You can see the size of that gap even in the category’s best story. Digit’s 65,000 paid hours across nine customer sites were all earned inside barrier-protected work cells; Agility’s fence-free next generation is only beginning its first customer deployments at the end of this year. And per its SEC filings, Agility spent roughly $111 million running the business in 2025 against net sales of $1.8 million. That is what the distance between a demo and a business costs, and it explains why the durable companies make quiet engineering choices early: they layer recurring software and service revenue over the machine, they design for minimal customer retrofitting, and they race to the second and third site.

So when we evaluate a robotics company, we try to grade it the way an operator would. Cost per unit of completed work matters more than cost per hour, the qualification we flagged at the outset, because a robot working at half a human’s pace effectively costs twice its wage. Renewal rates matter more than pilot counts. Time to the second site matters more than the quality of the demo. Those questions show up in what we own. Chef’s no-retrofit design is a deployment decision disguised as a hardware one, and Machina Labs, whose robotic craftsmen form sheet metal to millimeter precision and count Lockheed Martin Ventures and NVIDIA Ventures as co-investors, wins because it delivers finished parts rather than machinery.

5. Humanoids Are Real, Long-Dated, and Priced Ahead of Themselves

The question we hear most often is about humanoids, and it deserves a straight answer. Humanoid deals cross our investment committee regularly, so this is not a category we dismiss. It is a category we price, and we price it differently than the narrow applications in idea two, because the two are different assets. A tote-moving robot is a claim on wages being earned this quarter. A general-purpose humanoid is a claim on wages that may be earned next decade. The first is underwritten on renewal economics; the second on entry price and time.

The honest scorecard has two columns. The technology is advancing fast. TrendForce expects global humanoid shipments to jump roughly sevenfold this year to more than 50,000 units, and Digit proves a humanoid can earn wages today in a narrow lane. But most of those shipped units are research and light-duty machines, certification for side-by-side human work is only now reaching its first tests, and dexterity in messy, unstructured environments is still a research problem. The technology, in other words, is early. The prices are not. Figure raised at a valuation well ahead of its revenue, its secondaries now trade below that mark, and Unitree has demonstrated that the body itself can be built for $13,500. Paying tens of billions for a body-maker means believing both that the hardware won’t commoditize and that a price already set for perfection leaves room to be right. We believe neither.

Our answer is portfolio construction rather than abstinence. We own the transition through the wedges earning wages today and through the layers every humanoid will eventually need, we keep our entry prices early-stage, and we judge humanoid claims by renewal economics rather than choreography. General-purpose humanoids are real, and they are long-dated. Both things are true, and the second one is what late-stage prices ignore in our view.

What Would Change Our Mind

A thesis worth publishing is a thesis worth testing, so here are our tests in advance. We would revisit this view if Digit’s contracts renew poorly, or if Robots-as-a-Service pricing erodes faster than deployment costs fall, because either would suggest the wage line commoditizes before the moats form. We would revisit it if safety certification for working alongside humans slips well past 2028, because caged robots can only reach a fraction of the payroll this thesis depends on. We would revisit it if general-purpose robot brains get so good that one fleet’s data stops mattering, because that would weaken the flywheel we underwrite. And we would revisit it if labor markets slacken enough to refill the empty shifts, because the entire argument rests on zero output being the alternative. None of these has happened. All of them are checkable. We would rather publish the kill conditions now than explain them later.

Bottom Line

Five ideas, one breath. Robots are priced against wages now, so the market is payroll rather than equipment. The money today is in narrow applications, not general-purpose promises. The body is commoditizing, so returns are moving to software, perception, and data. The gap between a demo and a deployment is where companies live or die. And humanoids are real but long-dated, which makes entry price everything.

Or in one sentence: robots stopped being equipment and started being labor, so the market stopped being the world’s capex budgets and became the world’s payrolls. Investors still grading robotics against the machine-tool market will keep concluding it is too small, and they will keep being wrong for the same reason. They are reading the wrong line of the income statement.

You already invested in AI’s mind. It is getting a body, and the body just got a wage.

To discuss robotics and physical-AI exposure with an AV Senior Partner, or to explore AV’s deep tech and AI funds, visit AV.VC. The AV Syndicate is free to join, with deal-by-deal access and no obligation.

Sources

1. Digit deployments at Schaeffler (Cheraw, SC, eight-hour shifts since early 2025) and GXO (Flowery Branch, GA, 100,000+ totes): Agility Robotics, “Digit Moves Over 100,000 Totes in Commercial Deployment,” Nov. 2025; The Robot Report, “Humanoid maker Agility Robotics to go public through SPAC merger,” June 2026; deployment coverage, Schaeffler Tomorrow / Humanoids Daily, 2025–2026.
2. Digit operating cost of $10–$12/hour with path to $2–$3; modeled RaaS pricing of ~$8,500/month per robot; ~$30/hour fully loaded human labor comparator; payback under two years: Agility Robotics / Churchill Capital Corp XI investor presentation and SPAC communications, June–July 2026 (SEC Form 425).
3. Agility Robotics SPAC: ~$2.5 billion merger with Churchill Capital Corp XI (Nasdaq), announced June 24, 2026, ~$620 million expected gross proceeds: company announcement via Business Wire; GeekWire, June 24, 2026; The Robot Report, June 30, 2026.
4. Agility 2025 financials: ~$111 million operating expenses (up from ~$71 million in 2024), $1.8 million net sales, ~$100 million cash burn: Agility Robotics Form S-4 / SEC filings, as reported by The Robot Report, Sept. 2026, and GeekWire, June 2026.
5. Digit 65,000+ operating hours across nine customer sites; current deployments in barrier-equipped work cells; fence-free Digit v5 first deployments beginning Dec. 2026: Agility SEC filings and SPAC communications, 2026; Agility Digit v5 announcement, June 2026.
6. Toyota Motor Manufacturing Canada deployment (seven Digit robots), February 2026: Agility Robotics press release via Business Wire, Feb. 19, 2026; TechCrunch, Feb. 19, 2026.
7. 2026 robotics venture funding surpassing the full-year record by mid-year: Crunchbase News, “Sector Snapshot: Robotics Startups On Fire As Venture Funding Surges To Record Numbers In 2026.”
8. a16z $1.1 billion Machine Age Fund for the physical buildout of AI (chips, data centers, robotics): Andreessen Horowitz announcement, Aug. 28, 2026; TechCrunch, Aug. 28, 2026.
9. Unitree Robotics: Shanghai STAR Market debut closing up ~460% (Aug. 19, 2026); 5,500+ humanoids shipped in 2025; G1 flagship from ~$13,500; Q1 2026 adjusted net profit down 52.55% year-over-year: Unitree IPO prospectus; Bloomberg, Aug. 2026; The Robot Report, Sept. 2026; STAR Market Daily / SCMP coverage, May 2026.
10. Amazon one-millionth warehouse robot (July 2025); ~75% of deliveries robot-assisted: TechCrunch, July 1, 2025; CNBC, July 2, 2025 (per WSJ reporting cited therein).
11. S. manufacturing labor gap: 3.8 million workers needed by 2033, ~1.9 million potentially unfilled; 65% of manufacturers cite attracting/retaining talent as top challenge: Deloitte and The Manufacturing Institute, 2024 manufacturing workforce study; NAM Q1 2024 Manufacturers’ Outlook Survey.
12. Figure: $1B+ Series C at $39 billion post-money (Sept. 16, 2025), ~15x markup from Feb. 2024; secondary-market pricing below the Series C mark as of mid-2026: Figure press release via PR Newswire; TechCrunch, Sept. 16, 2025; Forge Global and Nasdaq Private Market indicative secondary prices, June–Aug. 2026.
13. Humanoid shakeout: K-Scale Labs shutdown (Nov. 2025, The Information); Cartwheel Robotics shutdown (Feb. 2026, Humanoids Daily); Sanctuary AI pivot to software (June 2026); Amazon acquisition of Fauna Robotics (March 2026): contemporaneous press coverage.
14. Defense & security robotics as largest 2025 segment: $8 billion across 234 deals: PitchBook, Q4 2025 Robotics & Physical AI VC Trends.
15. TrendForce forecast: 2026 global humanoid shipments to exceed 50,000 units, >700% year-over-year growth: TrendForce, “AI to Reshape the Global Technology Landscape in 2026,” Nov. 2025.
16. Robotics training-data landscape (no single data modality winning; deployment data accruing to fleet owners): proprietary market research reviewed by AV, summer 2026, based on 41 market-participant interviews (July–Aug. 2026).
17. AV portfolio company descriptions, valuations, co-investors, and the Ghost Robotics 2024 exit: AV portfolio database.

Portfolio companies shown are examples for illustrative purposes only, are not necessarily indicative of any AV fund or investor, and are not available to future investors, except potentially in the case of follow-on investment. Past performance does not guarantee future results.

This communication is from Alumni Ventures, a for-profit venture capital company that is not affiliated with or endorsed by any school. It is not personalized advice, and AV only provides advice to its client funds. This communication is neither an offer to sell, nor a solicitation of an offer to purchase, any security. Such offers are made only pursuant to the formal offering documents for the fund(s) concerned, and describe significant risks and other material information that should be carefully considered before investing. For additional information, please see here. Achievement of investment objectives, including any amount of investment return, cannot be guaranteed. Co-investors are shown for illustrative purposes only, do not reflect all organizations with which AV co-invests, and do not necessarily indicate future co-investors. Example portfolio companies shown are not available to future investors, except potentially in the case of follow-on investments. Venture capital investing involves substantial risk, including risk of loss of all capital invested. Diversification cannot prevent investment loss; it is a strategy to mitigate investment risk. This communication includes forward-looking statements, generally consisting of any statement pertaining to any issue other than historical fact, including without limitation predictions, financial projections, the anticipated results of the execution of any plan or strategy, the expectation or belief of the speaker, or other events or circumstances to exist in the future. Forward-looking statements are not representations of actual fact, depend on certain assumptions that may not be realized, and are not guaranteed to occur. Any forward-looking statements included in this communication speak only as of the date of the communication. AV and its affiliates disclaim any obligation to update, amend, or alter such forward-looking statements, whether due to subsequent events, new information, or otherwise.