$1.8M Revenue, $2.5B Valuation: What the Filings Reveal About Humanoid Robots

What it's about

Humanoid robots are touted as the next big AI market. CEOs talk about millions of machines in factories and warehouses. Until now, there were hardly any hard numbers to back that up, because most manufacturers are private and only publish what helps them. That's changing. Since November 5, 2026, Agility Robotics has traded on the Nasdaq under the ticker AGLT. It's the first pure-play humanoid company on a US exchange. To go public, Agility had to open its books to the US Securities and Exchange Commission (SEC). At the same time, we now have the IPO prospectus of Chinese market leader Unitree and the half-year report of Hong Kong-listed UBTech.

We put these mandatory disclosures side by side. Our thesis: so far, humanoids sell mainly as research and demo equipment, not as factory workers. Press releases also routinely blur three things you need to keep apart:

• Order: a letter of intent or a contract, often with conditions attached.
• Delivery: the robot has left the factory.
• Booked revenue: under accounting rules, the money is actually on the books.

Agility: $2.5 billion valuation, $1.8 million in revenue

Agility went public through a so-called SPAC merger with Churchill Capital Corp XI. A SPAC is an empty listed shell that merges with a private company. This skips the traditional IPO and its rigorous price discovery. According to the investor presentation, the pre-merger valuation was $2.5 billion. The June presentation projected about $620 million in fresh capital. In the end, Agility says it raised about $465 million gross: roughly $164 million from the SPAC's trust account, about $201 million from a private placement led by Foxconn, and $100 million that had already come in earlier. Redemptions explain the shortfall: about 61.5% of SPAC shareholders chose to take their cash back rather than invest in Agility.

The audited figures in the prospectus are sobering:

• 2025: $1.78 million in revenue. According to the financial statements, $1.13 million of that (about 64%) came from related parties, meaning investors or closely affiliated companies. More than half of the year's revenue came from five robots bought by an investor. Direct costs were $4.47 million, putting gross margin at roughly minus 151%. The net loss was $138.1 million.
• First half of 2026: $174,362 in revenue, down from $624,866 a year earlier. According to TechCrunch, most of it came from related parties. The net loss was $361.9 million. A large share of that likely comes from non-cash items such as stock-based compensation, but we weren't able to verify the exact breakdown in the original filing.
• Going-concern warning: Agility itself writes that without the merger, there would be “substantial doubt” about its ability to continue as a going concern.

Measured against its market value, Agility's pre-merger valuation is a little over 1,400 times its 2025 revenue. Agility doesn't disclose how many Digit robots it has delivered. What is documented: nine customer sites where Digit is deployed or firmly planned, and more than 65,000 operating hours. Named customers include GXO, Schaeffler, Toyota Motor Manufacturing Canada and Mercado Libre. Agility says its RoboFab factory in Salem, Oregon, could build 10,000 robots a year.

The $300 million order up close

The headline “more than $300 million in orders for Digit 5” carries the whole IPO story. The fine print tells a different story:

• It's a single customer and 1,000 robots under a three-year “Robots as a Service” (RaaS) rental model. RaaS means the customer pays a monthly fee while Agility keeps ownership of the robots. So Agility bears the upfront cost of the hardware.
• According to the amended filings, this customer is a related party. Exactly who remains undisclosed.
• Agility says the order is subject to “contractual milestones, features and specifications.” Agility itself writes that the amount is “not current-period revenue.”
• For every robot procured, the customer receives 453 warrants, meaning the right to buy shares later at a fixed price. For 1,000 robots, that's 453,000. According to an analysis at Invest Deep Tech, Agility was still working out in July how to account for this. Sweeteners like these can push recognizable revenue below the $300 million.
• Digit 5 was only unveiled on September 15, 2026. First customers are supposed to get it in the first half of 2027, with general availability planned for the end of 2027.

Do the math: $300 million for 1,000 robots over three years works out to about $100,000 per robot per year. That matches the monthly rate of roughly $8,500 in the investor presentation. If the order is fully delivered, it would mean about $100 million in annual revenue, a good 55 times the 2025 figure. Our take: this is a target with strings attached, not a backlog in the usual sense.

Unitree: lots of robots, but mostly for labs

Unitree has been listed on the STAR Market, Shanghai's tech board, since August 19, 2026. The offering price was 150.80 yuan, and the stock opened at 1,100 yuan. Unlike Agility, Unitree makes money: about 1.7 billion yuan in revenue in 2025 and about 600 million yuan in adjusted profit.

The unit numbers show the problem with the three terms:

868 million yuan divided by 5,215 units comes to about 166,000 yuan per robot, or just under $25,000. Unitree told Caixin that it counts only humanoids delivered to end customers. The industry has no binding definition. According to NextMSC, the industry association IFR counts only about 7,000 humanoids sold worldwide in 2025, while Omdia counts 13,318. The difference comes mostly down to what each one includes.

The most revealing number is the breakdown by customer group that the prospectus gives for the first nine months of 2025:

• Research and education: 73.6%
• Commercial use (demos, exhibitions, events): 17.39%
• Industrial applications: 9.01%

Even that 9% is a generous count. According to Sina Finance's analysis of the prospectus, 50 to 70% of it is reception and guide robots at companies. Actual industrial use cases brought in only 15.7 million yuan. With humanoid revenue of 595 million yuan in that period, by our calculation that's about 2.6%. In the third quarter of 2026, Unitree's revenue rose 70% to 665 million yuan, but profit fell 17%. The quarterly report doesn't break revenue down by product.

UBTech: orders from data centers, continued losses

Shenzhen-based UBTech is listed in Hong Kong and reports every six months. For 2025, the company reported 821 million yuan in revenue from full-size humanoids, 1,079 units sold and a net loss of 790 million yuan. Orders for the Walker series added up to almost 1.4 billion yuan by the end of December 2025. Many of the large individual orders came from humanoid data collection and training centers, for example in Jiujiang (143 million yuan), Zigong (159 million), Huizhou (60 million) and Hohhot (78 million). At these centers, robots practice movements to generate AI training data. They aren't working on a production line.

In the first half of 2026, UBTech sold 921 full-size humanoids for 590.3 million yuan. Besides the Walker S2 industrial robot, this category includes other models. That works out to about 640,000 yuan per unit, almost four times Unitree's figure, probably because solutions and project business are booked in as well. The net loss was 338.8 million yuan. The 2026 target is 10,000 full-size units, so one half-year delivered just under a tenth of that.

Figure, Tesla, AGIBOT: announcements without financials

The best-known US names don't have comparable mandatory disclosures. According to NextMSC, Figure had built more than 350 Figure 03 robots by the end of April 2026. They went mainly into internal development, data collection and pilot projects. At BMW in Spartanburg, Figure 02 robots worked for about eleven months and helped build more than 30,000 BMW X3s. Tesla said on its third-quarter earnings call that it has started “early production” of Optimus in Fremont. Tesla hasn't given a unit count, and external customers are supposed to get deliveries starting around the end of 2027. Omdia ranked AGIBOT first for 2025 with 5,168 units. It has no prospectus of its own, but the company has taken control of the listed materials maker Swancor.

Our verdict

The mandatory disclosures largely confirm our thesis. The only maker selling humanoids in meaningful numbers and turning a profit is Unitree, and mostly with low-cost units for universities and labs. UBTech books high revenue from data centers and keeps losing money. Agility can point to years of real-world deployments in warehouses but has barely any revenue. Its most important order comes from a related party and has conditions attached.

Our take: whenever you read news about humanoids, ask which of the three numbers is meant. Is it a conditional order, a delivery or booked revenue? And who's the customer: a factory, a lab or an investor? These are exactly the distinctions that get lost in many press releases. To be fair, Agility is the first US company that has had to disclose all its numbers, which competitors from Figure to Tesla have not yet done. Whether Digit 5 hits its promised throughput, runtime and accuracy won't be clear until 2027 at the earliest. At press time, it wasn't yet known where AGLT stock closed on its first trading day.