Humanoid Robots Have a New Proof Metric: What 2026 Deployments Actually Show

Figure AI operates more robots than it employs people. Agility Robotics reports $1.8 million in 2025 revenue and a $140 million operating loss. Both facts are true. Neither appears in a press release about backflips.

Read that opening again. The gap between headline and evidence is now the sector's defining metric, and it is widening exactly where the money is flowing.

In 2025, humanoid robotics pulled in a record $13.8 billion globally, up from $7.8 billion the year before, per Embodied Global's funding report. Goldman Sachs projects a $38 billion market by 2035. Tesla is building a dedicated Optimus production line in Fremont. Neura Robotics closed a $1.4 billion Series C in June. Apptronik sits at roughly $935 million raised. Figure AI carries a private valuation near $39 billion.

What a Verified Deployment Looks Like

Start with the cleanest number in the entire sector: BMW Group states that a Figure 02 robot supported production of more than 30,000 X3 vehicles at its Plant Spartanburg over ten months, moving more than 90,000 components in around 1,250 operating hours. That is operator-side confirmation: the customer, not the robot maker, published the figures. The automaker has since expanded the program, bringing Figure 03 into Hall 52 for a sequencing workflow under Helix 02 control.

That single deployment matters more than every funding round combined, because it answers the question investors actually ask: does this machine work in a production environment for months at a time? Yes. One of them does.

Why Nobody Trusts the Other 43%

The sector's public record is overwhelmingly vendor-authored. A robot company saying its robot works is a marketing claim. A plant manager saying it works, with shift hours attached, is evidence. That distinction is the entire investment thesis of the humanoid sector in 2026, and it is not settled.

Axis Intelligence, an independent research firm, built the Deployment Verification Ratio to quantify the gap. Its tracker documents seven deployments as of August 9, 2026. Four pass the operator-side test. One deployment in the entire world, the automaker's concluded Figure 02 run, has published hours, parts and output. The rest rest on vendor disclosure alone.

The Two Companies With Actual Numbers

Agility Robotics has the most documented commercial deployment in the sector. Its June 2026 SEC-filed investor presentation reported 65,000 hours of operation across nine committed customer facility deployments as of May 2026, naming Schaeffler, GXO, Amazon, Toyota Motor Manufacturing Canada and Mercado Libre as Digit customers. The GXO tote workflow ran at 98% accuracy and moved more than 100,000 totes. Mercado Libre's workflow hit 98% accuracy on roughly 25,000 totes. That is a vendor-filed account, not operator confirmation.

But note what these are: numbers from a vendor's own SEC filing. The company states that Digit v4 carries up to 35 pounds, that its bill of materials is approximately $125,000, and that RoboFab in Salem, Oregon is designed for 10,000 units of annual capacity. None of that is independently confirmed. The one site-level record inside the filing, Digit v4 at Schaeffler's plant in Cheraw, South Carolina, placing 25-pound baskets into an industrial washing machine, is evidence tier T3: vendor disclosure in a regulated filing, with no operator account from Schaeffler.

The Comparison Table Investors Need

Read the table vertically. The highest valuations sit on the least independent evidence. Tesla's implied standalone value of $150 billion or more rests on zero published Optimus deployment counts. Axis Intelligence excludes the company from its tracker entirely because no Tesla document states units at a named site. The widely quoted "50,000 cumulative Optimus units" traces to no filing.

Tesla's Black Box Is the Whole Market's Problem

Elon Musk has said scaling Optimus will be harder than any manufacturing program the company has attempted. Tesla cleared the Fremont Model S and X line in 46 days to make room for the first Optimus production line, and Musk walked the site on July 1 saying production had not begun and initial output would be extremely slow, citing roughly 10,000 unique parts.

There is no Optimus order page. There is no price. There is no delivery commitment. The Q2 2026 earnings call was the next dated checkpoint, and Q1 guidance pointed to Fremont production beginning late July or August 2026. A company claim, like everything else about the program.

That is the paradox in its purest form: the company that may control the largest share of this market's future value is also the company with the least verifiable current evidence. You cannot model it. You cannot discount it. You can only hold it against the sector's one honest benchmark, operator-confirmed deployment hours, and note that on that benchmark, Tesla does not appear.

Where the Money Went

The capital structure reveals the sector's real division. US and European companies raise on software and foundation-model story. Physical AI is the pitch, hardware is the constraint. Neura's round was milestone-contingent, tying disbursement to production and deployment targets, a structure that signals even sophisticated backers are hedging against the evidence gap. Meanwhile China's Galactic Universal shipped 5,168 humanoids in 2025 and targets 20,000-plus in 2026, volume at a price point Western firms cannot match.

And then there is the corporate-backer pattern: BMW, Mercedes-Benz, Schaeffler, Foxconn and Amazon are not just investors, they are deployment sites. That is the sector's quiet strength. The customers with the factories are also the shareholders with the balance sheets, and they keep buying robots after the pilots.

The First Public Test of Humanoid Credibility

Agility's SPAC merger with Churchill Capital Corp XI is the first pure-play humanoid robot company to pursue a US public listing. The deal values the company at roughly $2.5 billion, with about $620 million in gross proceeds: approximately $420 million from Churchill's cash trust and a $200 million PIPE at $10 per share led by Foxconn. The ticker is AGLT. The S-4 registration was confidentially submitted in July.

The filing that matters is the one that emerged before the SPAC: $1.8 million in 2025 revenue against a $140 million operating loss. The Robot Report surfaced those numbers on September 7, and they are the first real financial disclosure the humanoid sector has produced in public markets.

Whatever multiple public investors assign to its revenue and order pipeline becomes a reference point that reprices Figure, 1X Technologies, Apptronik and every private humanoid company by comparison. The sector has spent two years raising on 2028–2030 assumptions. AGLT is the first time a market price touches current revenue.

What the Buyers Actually Do

Enterprise buyers are deploying, not piloting. The language has shifted even where the numbers have not. Schaeffler plans to deploy a four-digit number of Humanoid robots across its facilities by 2032 while supplying more than half the actuators for the wheeled platforms through 2031. The same company agreed to deploy at least 1,000 Hexagon Robotics AEON humanoids across its factories over seven years. Apptronik's Apollo runs at Mercedes-Benz, GXO Logistics and Jabil, and Google designated Apollo the exclusive hardware platform for its Gemini Robotics project.

The industrial supply chain is building itself around the category before the category has proven itself. The German industrial group is simultaneously the customer, the component supplier and the data collector. That is not a hype position. It is an industrial-hedge position: buy the option on humanoids cheaply, in the form of actuators and pilot deployments, while the evidence base matures.

The Quiet Build-Out: Safety Certification Infrastructure

The least discussed signal in the sector is not a robot at all. It is the emergence of a certification stack designed for machines that work beside people without barriers. NVIDIA launched what it called the industry's first full-stack safety architecture for humanoid robots and named the Oregon company its inaugural partner. Certification runs through the NVIDIA Halos AI Systems Inspection Lab, which is accredited by the ANSI National Accreditation Board and recognized by TÜV Rheinland, UL Solutions, TÜV SÜD, exida, SGS and CertX. The $300 million order book for 1,000 Digit v5 units, disclosed at the Fremont facility announcement, is the largest quantified fleet commitment in any filing.

Digit v5's specifications were chosen with this in mind: the robot stands about 5 feet 9 inches, lifts up to 50 pounds and operates up to 22 hours per day. The 50-pound limit is aligned with OSHA's recommended maximum for human manual lifting, a number deliberately picked to signal that Digit fits within human workplace frameworks rather than replaces them. Its safety software and AI components will be pre-assessed against IEC 61508, ISO 13849 and ISO/IEC TR 5469 before final third-party certification.

This matters for investors because certification is the gate that separates demonstrations from repeatable revenue. A robot that cannot pass a recognized safety standard cannot be deployed at scale in a regulated industrial environment, and every deployment after the pilot depends on it. The first companies to stand up an auditable certification path, not just a safety marketing page, get the compounding advantage, because every new customer contract cites the certified baseline rather than starting from zero.

It is also why the sector's silent divergence is significant: Chinese volume players ship thousands of units a year, but Western certification frameworks are being written around Western platforms. The race is no longer only about who builds the best robot. It is about whose robots are legally deployable in the factories that have the purchasing power.

Why the Home Robot Story Is Still a Lie

The 88% household task failure rate documented in 2025–2026 research still applies to general-purpose home use. Outside automotive, warehousing and factory showcases, humanoid deployments are experimental pilots.— RoboBrief, Humanoid Robots in 2026: The Complete Tracker

1X publishes NEO order terms: $20,000 to own or $499 per month, with a $200 refundable deposit and US deliveries stated to begin in 2026. Unitree lists the H2 Plus at $100,000 on its official store but marks it unavailable and directs buyers to contact sales. Both are real commercial offers. Neither is evidence that the home market works, and the research on household task reliability says it does not yet.

RoboZaps, which tracks verified status, categorizes the field into five evidence levels: 1X has explicit public order terms, Unitree has a sales-assisted catalog listing, Figure has a named factory workflow, the Fremont company is preparing production, Atlas has committed enterprise deployments. Five different levels of commercial certainty. The useful news is the change in status, not the headline.

No one in the sector publishes mean time between intervention, how long a humanoid runs before a human has to step in. That number, when someone publishes it, will be the most honest figure in the field. Watch for it before you watch for the next funding round.

And watch the arbitrage: Boston Dynamics said in January that all 2026 Atlas deployments were already committed, with fleets scheduled for Hyundai's Robotics Metaplant and Google DeepMind. A company choosing its deployments over the next eighteen months is a company with more demand than supply. That is a signal of a different kind: the ones with operator commitments are the ones not publishing them yet.

The sector is no longer science fiction, and it is not yet an asset class with reliable financials. It sits in the gap between the two, and for the first time, a public market is about to price that gap. As we wrote in August about Persona AI's bet on humanoids that weld ships nobody can staff, the deployment question was always going to be the real test. The difference now: the test has a stock ticker.

Sources