Executive Summary

Key Numbers

Sources: the Agility Robotics and Churchill Capital Corp XI merger registration statement (September 2026) and reporting by TechCrunch and The Robot Report (November 5, 2026)

Agility Robotics is based in Oregon, and it builds Digit, a two-legged robot that moves boxes of goods around warehouses and factories. It arrived on the public market through a merger with a blank-check company rather than a conventional IPO. The merger with Churchill Capital Corp XI closed on November 4, and the first trade under AGLT went through on Nasdaq the next day. The equity value set just before the merger was $2.5 billion.

The cash the company actually took in is about $465 million. Of that, $164 million came from the blank-check company's trust account, $201 million from a pre-arranged investment led by Foxconn at $10 a share, and $100 million from financing that arrived right after the merger was announced. Nvidia, Amazon, SoftBank Vision Fund 2 and the German component maker Schaeffler were already shareholders, and three Korean firms, Palo Alto Capital, SW Investment and Elohim Partners, joined the pre-arranged round.

As recently as June, the deal was expected to bring in more than $620 million. The gap comes from redemptions. A shareholder in a blank-check company who does not want the merger can take their money back and leave, and this time about 61.5 percent of common shareholders chose that. The trust account fell from $424 million to $164 million, and the shortfall came straight out of the proceeds.

The first day of trading was subdued. The stock opened at $11.26, dipped to $8.40 at one point, recovered to $11.49, and closed around $11, briefly below the $10 the pre-arranged investors paid. Set next to Unitree, which jumped hard on its first day in Shanghai in August, the difference in temperature is plain. The market did not price the word humanoid on its own.

The $2.5 billion also reads differently alongside the record behind it. The Series C that raised $400 million in 2025 valued the company at about $2.12 billion, so the step up over roughly a year is modest. Peers that stayed private carry larger numbers over the same stretch: Figure AI at about $39 billion, Neura Robotics at about $7 billion, Apptronik at about $5.5 billion. Those values are set a different way and do not compare directly, but it is at least visible that reaching the public market first does not mean carrying the highest price.

The merger registration statement published in September spells out how this company earns money today. Revenue in 2025 was $1.782 million, up 474 percent from $310,000 the year before, while cost of revenue over the same period was $4.473 million. Every robot that went out the door lost money. Operating loss was $140.17 million and net loss $138.1 million, more than double the prior year. The filing also carries going-concern language, saying the company lacks the working capital to last a year without the proceeds of this deal. Revenue in the first half of 2026 fell further, to $174,000, and most of even that came from related-party transactions. More than half of the 2025 revenue came from a single investor buying five robots, Korean outlets report.

The order book has a large number on it. There is a $300 million contract to supply 1,000 Digit 5 units. The buyer is a single party, the contract is a three-year subscription-style supply deal, and it stays in force only if contractual milestones and product specifications are met. The warrants attached to it vest in proportion to robots actually deployed. The filing itself states that the amount is not a measure of current-period revenue. Digit 5 shows as a prototype at CES in early 2027 and reaches general availability only at the end of that year. Beyond that, what the company has disclosed about its pipeline is that it is in discussions with about 30 prospective customers and aims to have 100 by the end of 2027.

So there was no present-tense number to put in front of investors. What the filing put in that place is the 65,000 hours Digit v4 has logged at customer sites. The figure is cumulative through May 2026, and the same number appeared in an investor deck that June. Digit is deployed at nine customers today, among them the logistics company GXO, Schaeffler, Toyota Motor Manufacturing Canada, and the Latin American e-commerce firm Mercado Libre. At GXO alone it has moved more than 100,000 totes.

What Digit v4 does on site is move boxes. The specification the company has given is repeated handling of loads up to 50 pounds, about 23 kilograms, at 98 percent accuracy. Manufacturing runs through RoboFab, the robot plant the company built in Oregon. Once fully equipped it can produce up to 10,000 units a year and sources about 75 percent of its parts inside the United States, the company says.

Chief executive Peggy Johnson calls those hours a data flywheel. Digit may look slow next to the demo videos from Figure, Tesla, 1X and Boston Dynamics, her argument runs, but it is doing real work and earning money, and the field record piling up that way will build the next generation faster. Operating hours have taken the place of units sold as the claim to advantage.

Revenue has accounting rules. When it can be recognized, what gets added and what gets left out, all of it is defined, and an auditor checks it separately. Operating hours have nothing of the kind. The 65,000 hours is a figure the company counted by its own standard and wrote into its own filing, and that standard was not written down alongside it.

Start with the empty boxes and there is no denominator. How many robots shared the total is not public, so there is no telling whether one machine ran a long time or many machines each added a little. There is no breakdown by site, by customer, by task, or by software version. Hours powered on, hours available for work, hours actually moving, and hours doing productive work are not separated.

The largest empty box is the human share. How much teleoperation, human intervention, resetting, error recovery and over-the-shoulder supervision is mixed into the number was not disclosed. Without separating hours worked autonomously from hours a person was at the controls, "it ran in the field" is not evidence of capability. There is no maintenance, parts-replacement or mean-time-to-repair record on the failure side either, and no third-party verification that a customer confirmed the total.

None of this says the number is false. That Digit is at nine real sites moving boxes is confirmed separately. What the figure proves stops at activity having taken place. Using it as a fleet-level performance measure would take a denominator, a breakdown and verification arriving with it, and right now the filing holds only the one-line total.

The 65,000 hours piled up in other people's warehouses and factories, not in an Agility lab. They are what the sensors recorded while Digit moved through GXO's distribution centers, Schaeffler's plants and Toyota's assembly lines. Inside that record are the customer's work paths, process layouts and swings in throughput. To the robot company it is training data; to the customer it is closer to a blueprint of their own plant.

So where the rights to this data go starts to matter, and the filing leaves that part blank. The risk factors cover infringement of intellectual property, the difficulty of enforcing rights abroad, and workplace safety. A clause on who owns data collected at customer facilities is not among them. The company's own materials explain only that real operating time produces proprietary deployment data, and say nothing about how that exclusivity is secured by contract.

From the investor's side the gap means the basis of the asset is unsteady. If data is the moat, whether the moat belongs to the company or has to go back when a customer ends the contract is a premise of the valuation. From the customer's side the question runs the other way. If the record gathered in our plant teaches the next robot, and that robot is also sold to our competitors, what is it that we handed over?

Putting data forward as an asset without filing the deed alongside it is not unusual in physical AI right now. There is no shared standard for counting the hours a robot worked, and no standard contract that settles ownership of the data collected that way. What makes this listing conspicuous is that the gap has landed on a public-market document for the first time.

Two days before Agility's Nasdaq debut, on November 3, Korea's first AI Humanoid Data Factory opened in Yuseong-gu, Daejeon. The Ministry of Science and ICT, the Korea Institute of Machinery and Materials and ETRI run it together, at a cost of about 105 billion won across five years from May 2025 to April 2030. Twelve humanoids are running now, rising to 25 by 2030. Nine Korean universities including KAIST and GIST and five overseas research institutes take part. A facility like this was built on the public side first because industry had asked for it, saying no single company can gather the manufacturing and logistics data needed to teach a humanoid on its own. The ministry has also been pushing a national manufacturing data library since July 2026.

The way the data is gathered differs from Agility's. In Daejeon a person drives the robot directly over a remote link and the joint values are recorded, or a person demonstrates a motion and the robot imitates it while human and robot movements are stored as matched pairs. It is data that people make. What Agility put forward is time a robot spent running in a real workplace. The two approaches build records of a different character, and the questions that follow differ with them.

The timing happens to overlap as well. Daejeon starts opening its data to industry at the end of 2027, and Digit 5 reaches general availability at the end of 2027. What this blog pointed out on opening day, in At Korea's AI Humanoid Data Factory, Robots Learn Housework First, was that the data format, the transfer terms, the scope of commercial use and the eligibility to apply had not been published. The box left empty at the public facility and the box left empty in the listing document are the same box.

Another axis is taking shape in Korea. Boston Dynamics, part of Hyundai Motor Group, is preparing to list in 2027, and Korean outlets have read Agility's debut as one reference point for it. The two companies work in different fields with different customers and a different revenue base, though, so Agility's share price does not carry over directly. What does carry over is the question rather than the number. What has the public market started to demand from a robot company as evidence?

Writing data down as an asset has been talk for a long time. This time it arrived on a securities document as a single number, and with that, asking how the number was made became part of an investment judgment. How the hours a robot worked get counted, and what rules settle who owns the data inside those hours, will set the price of this asset from here.

What that takes, in the end, is lineage for the data. Each record has to carry which robot it came from, which site, which software version, and whether the robot was working autonomously or a person was at the controls, before the total can be taken apart. The same record has to carry which customer's facility it came out of and how far the contract with that customer allowed reuse, before rights can be told apart. Data without those two lines can be collected, but it is hard to count as an asset.

If you work anywhere near data coming off robots or plant equipment, there are three things worth getting in writing the moment the other side starts calling that data an asset.

• What the unit of counting is. If it is counted in hours, is it defined as hours powered on, hours available for work, or hours actually working?
• Whether the human share stays in the record. Are teleoperation, intervention and recovery stored separately from autonomous operation?
• How far the other side may use records produced at our site. Does the contract distinguish model training from supply to other customers?

None of the three asks about the robot's performance. What they ask about is the structure of the record. Deciding at the start what gets recorded is what lets you show a basis later, when the record is called an asset, and leaving it undecided at the start means there is no way back once the records have piled up.

Thanks for reading this far. The circumstances of the listing day are covered at TechCrunch and The Robot Report, and the financial figures are laid out in GeekWire's reading of the filings. If you have ever had to write ownership and reuse scope for field robot or equipment data into a contract, we would like to hear how that sentence came out.

News Coverage

• 1.Fernholz, T. (2026). "Agility is the first pure humanoid robot company on the stock market. Are you buying?" TechCrunch, Nov. 5, 2026.
• 2.Crowe, S. (2026). "Agility Robotics begins trading on Nasdaq." The Robot Report, Nov. 5, 2026.
• 3.Bishop, T. (2026). "Agility's new Digit 5 robot lifts 50 pounds, works 20+ hours a day and operates alongside people." GeekWire, Sep. 15, 2026.
• 4.HuffPost Korea. (2026). "과기정통부 국산 로봇 '카이로스' 활약 앞당긴다: 국내 최초 휴머노이드 학습 전용 데이터 팩토리 공개" [Korean]. Nov. 3, 2026.

Analysis

• 5.Physical AI Guide Editorial Team. (2026). "Agility Robotics S-4: What Digit's Operating Hours, Pricing, Orders, and Losses Show." Physical AI Guide, Sep. 9, 2026.

Official Filing

• 6.U.S. Securities and Exchange Commission, EDGAR. (2026). "Form S-4 Registration Statement — Churchill Capital Corp XI / Agility Robotics, Inc. Business Combination." SEC EDGAR, Accession No. 0001213900-26-097764, Sep. 4, 2026.