$8.5 billion over 20,000 installed units is $425,000 each. Most of those units are robot arms, mobile robots and line integration. Not humanoids.
On November 12 in Munich, Agile Robots announced a $500 million Series F at an $8.5 billion valuation, led by SoftBank Group, with Nvidia, Hitachi and Hitachi Ventures, and T.Capital. AFP put it at €7.7 billion.
The most valuable line in the release is not the valuation. It is "brings the company's total invested capital to $2 billion."
Does the $2 billion include this round
"Brings to" points forward: the total after the money lands. The $500 million is inside, so the figure before was $1.5 billion.
Public numbers fit. A $220 million Series C in September 2021 pushed the valuation past $1 billion and made it Germany's first robotics unicorn. By 2022 tracked funding was $380 million; mid-2026 coverage put it near $1.5 billion.
15 plus 5 is 20. Three numbers that mesh.
What it actually sells
The release calls Agile a Physical AI company: humanoid and mobile robots, arms and hands, and the software behind them. Customers in automotive, consumer electronics, healthcare and services.
The assets: spun out of the German Aerospace Center, DLR, in 2018, starting in force control; 20,000 units installed; more than 2,000 in R&D; over 650 patents; a 20,000 square meter lab; 3,500 people from nearly 60 countries.
Force control is the old business: Diana 7, the flagship arm, carries torque sensors on all seven axes, turning "how hard" into a programmable quantity. The hardest assembly work is not reaching the position but what follows, how hard to press, how tight to turn, how deep to insert. Pure position control cannot do that, which is why the company holds ground in automotive and consumer-electronics assembly.
Then the purchases. In 2023 it took over the bankrupt Franka Emika, now Franka Robotics, one of the most common collaborative arms in embodied AI labs. The same year it bought a majority of idealworks, a BMW mobile-robot spin-out. In November 2025 it agreed to buy thyssenkrupp's automation engineering business in Europe and North America, closed April 1, 2026, renamed Krause Automation, with about 650 engineers and 10 sites.
That explains half the valuation. Not a team grown from zero, but an industrial automation group assembled over eight years with a Physical AI sign on the door.
The other half has to come from revenue.
€300 million, and 26 times
Chen Zhaopeng told the Wall Street Journal that 2025 revenue was about €300 million, that 2026 would roughly double, and that profit would come in two to three years.
At the AFP rate that is about $330 million. 8.5 over 0.33 is 25.7 times sales. A double to €600 million, about $660 million, brings it to 12.9.
Ordinary for software; for a seller of arms and integration it needs explaining.
The question is where the €300 million comes from. The asset list answers: arms, mobile robots, line integration, and 650 integration engineers just bought in. The humanoid line shipped its first generation, Agile ONE, only in November 2025, with production starting outside Munich in early 2026. The release gives no revenue for it, names four sectors and no customers.
Capacity is not shipment
Fürstenfeldbruck is described as having annual capacity for 10,000 legged and wheeled humanoids.
20,000 units over 8 years is 2,500 a year. One new plant's capacity is 4 times the average installation rate of the past eight years.
Not a boast necessarily: capacity is built for after 2027, and building before orders is routine in heavy industry. But the 10,000 is design capability, not orders or deliveries. Filling the line means selling 10,000 humanoids a year in a business with no revenue this year.
So part of the valuation buys an assumption: that the line gets filled.
At 3 times sales, a multiple I picked, the old business is €900 million, about $1 billion. Subtract from $8.5 billion and $7.5 billion remains, or 88%.
Nearly nine tenths of the valuation buys a line that is not full, not the 20,000 units already on customer floors.
Competition gets one line and no customer names. AFP added that European makers face Chinese rivals such as Unitree, whose edge is government support and price. Where the humanoid is priced decides whether the line fills.
Underneath sits a messier problem: bought revenue is low margin, and the 650 engineers lift revenue while pulling down blended margin.
What Nvidia and Hitachi are buying
SoftBank is financial logic; the other two are industrial. Nvidia is buying a position. Agile ONE SN ships with Isaac GR00T and Jetson Thor preinstalled, making Nvidia's stack a factory default. Franka arms already run in university and embodied AI labs, whose papers and data land on someone's hardware.
Hitachi is buying customers. Its manufacturing and energy clients on the operational technology side buy exactly force-controlled assembly and line integration, and adding a company with European OEM case studies to the catalog is cheaper than building it.
SoftBank's logic is simplest: it led the 2021 Series C and the 2026 Series F. Same firm, same target, five years, $1 billion to $8.5 billion.
Which number decides whether it is worth it
The release closes with Chen: Physical AI will fundamentally change how the world builds, moves and manufactures.
Unarguable, and untestable.
Three numbers can be tested: humanoids shipped in 2027, Krause's 650 engineers in consolidated revenue, and force control running inside an automaker's takt time without a person standing by.
The third is hardest and most valuable. It will not appear in a press release. It will appear on the customer's next line acceptance form.
$8.5 billion does not buy 20,000 arms. It buys a ticket to the next workstation.
*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*
