Nearly $11.7 billion in disclosed capital has gone into humanoid robotics. Not one of the best-funded companies in the category has a humanoid you can buy. The companies that do are the ones that raised the least.
The Humanoid Robotics Funding Race: What the Money Tells Us About Commercial Readiness
Lars Talbert, Founder | The Robotic Life
The Empty Quadrant
Plot capital raised against deployment stage for every humanoid company with a disclosed funding figure, and one region of the chart comes up empty. There is no company that has both raised serious money and put a humanoid into general commercial availability. The data comes from The Robotic Life’s Humanoid Robotics Funding Tracker and Companies index, as of 1 September 2026.
The numbers behind that gap are stark. Six companies have raised more than $500 million each. Together they hold $7.6 billion — roughly 65 percent of all disclosed capital in the sector. Every one of them is at pilot or prototype stage.
Four companies have humanoids in general commercial availability. Together they have raised $1.03 billion, under 9 percent of the sector total. The largest of them raised $340 million.
That is the finding worth sitting with. This is not the familiar observation that money and progress are imperfectly correlated. It is that at this moment, in this category, they are running on entirely separate tracks — and the reasons are engineering reasons, not financing ones.

Capital raised versus humanoid platform stage, 26 companies with disclosed funding, September 2026. Source: The Robotic Life research.
A Note on Method
The $11.7 billion total counts disclosed capital raised by humanoid robotics companies, including pre-IPO rounds for companies that have since listed. It excludes two things deliberately.
Internally funded corporate programs are counted separately. Tesla has disclosed more than $20 billion in dedicated capital expenditure and Xiaomi more than $10 billion in committed R&D across its EV and AI divisions. That is more than $30 billion, and it dwarfs everything else in the category — but it is a company deciding to spend its own money, not capital raised into a humanoid business. Folding it in would swamp the figure and make cross-company comparison meaningless.
Companies with undisclosed funding are excluded rather than estimated. Toyota Research Institute, Engineered Arts, Kawada, LG and several others publish no figure. Guessing would corrupt the total.
Stage refers to the humanoid platform only. Several companies here ship other robots commercially — Boston Dynamics’ Spot, Pudu’s service robots, JAKA’s cobots — while their humanoid sits at an earlier stage. The test for commercial is narrow and deliberately so: can a buyer place an order and receive a unit? Pilot means units are operating at sites the company does not own. Prototype means working hardware exists but has not left the building.
Where the Money Is
Capital is concentrated but less extremely than the headlines suggest. The five largest positions account for roughly 59 percent of disclosed funding. The concentration splits into three lanes.
The American mega-round lane
Figure AI leads all private fundraising with $2.34 billion in cumulative capital at a $39 billion valuation, following a Series C exceeding $1 billion in September 2025. Apptronik holds $1.1 billion at a $5.5 billion valuation, with Mercedes-Benz and Google backing the Apollo program. Both are at pilot stage.
The European counterweight
NEURA Robotics holds $1.6 billion in total funding at a $7 billion valuation, following a Series C announced in June 2026 as up to $1.4 billion — a figure contingent on the company meeting stated performance milestones, and worth reading as a ceiling rather than a settled balance. The round was led by Tether, with NVIDIA, Amazon, Qualcomm, Bosch and Schaeffler participating. The 4NE-1 platform is at prototype stage.
The Asian lane, and the exception inside it
UBTECH raised more than $1 billion privately before listing in Hong Kong in December 2023, and has Walker S units operating at BYD, Foxconn, Zeekr and Audi FAW plants. XPeng Robotics raised more than $900 million at a valuation above $6.3 billion. Galbot holds $635 million at $3 billion.
And then Unitree, which is the exception that makes the whole argument. Unitree reached commercial shipping and a public listing on roughly $155 million in pre-IPO private capital. Its G1 sells to anyone with a purchase order. It listed on the Shanghai Stock Exchange in August 2026. It raised about one fifteenth of what Figure has raised, and it is further along the commercial path than any Western company in the category.
The Four That Ship
The commercially available column is short: Unitree, EngineAI, LimX Dynamics and Rainbow Robotics. Three are Chinese; Rainbow is a Samsung subsidiary listed in Korea. Their combined disclosed capital is $1.03 billion.
None of them raised a mega-round. EngineAI, the best-funded of the four, holds $340 million. LimX has $296 million and has filed for a Hong Kong listing. Rainbow holds $240 million.
What they share is not a technical breakthrough the others lack. It is a different product decision. These are units sold to buyers who want a capable platform — researchers, integrators, developers, factories willing to do their own integration work. They are not selling a general-purpose worker that arrives ready to replace a job. That is a narrower promise, and a shippable one.
The companies holding the large rounds are pursuing the broader promise. That is a defensible strategy with a longer horizon. But it is worth naming plainly: the reason $7.6 billion sits with companies that have not shipped is not that they are behind. It is that they are attempting something harder.
The Readiness Ledger
Deployment status reorders the field completely, and pilot stage covers a very wide range.
Agility Robotics has the deepest operational record in the category. Digit has logged more than 65,000 operational hours per the company’s merger announcement, with enterprise customers including Schaeffler, Toyota Motor Manufacturing Canada, GXO and Mercado Libre, more than $300 million in multi-year orders for the next-generation Digit V5 pending contractual milestones, and over 30 customers in its pipeline.
That figure is worth reading as an engineering metric rather than a marketing one. Cumulative hours across live customer sites is the closest thing this industry has to a reliability dataset: it is where mean time between failures, wear under real duty cycles and the actual frequency of human intervention get measured. Laboratory demonstrations do not generate that data. Only deployments do, and it compounds — the fleet with the most hours has the most failure modes already catalogued and engineered out.
Agility reached that record on approximately $180 million in pre-SPAC private capital, under 8 percent of what Figure has raised. It signed a definitive merger agreement with Churchill Capital Corp XI in June 2026 at a $2.5 billion pre-money equity value, targeting more than $620 million in gross proceeds and a listing as AGLT. Per the company’s description, it would be the first US-listed pure-play humanoid robotics firm with active commercial deployments.
It still sits at pilot stage on this chart, because you cannot order a Digit. That is not a criticism of Agility — it is a description of how enterprise robotics is sold, and it is why pilot stage spans everything from a single unit on trial to a fleet with 65,000 hours behind it.
Three Filters Between a Funded Program and a Commercial One
Capital solves some problems and defers others. Three filters stand between a funded humanoid program and a deployed one. None is primarily financial.
Filter one: manipulation, not locomotion
Locomotion is substantially solved. The hard problem is contact-rich manipulation in unstructured environments, and the difficulty is data, not mechanics.
Locomotion generalizes well from simulation. A walking gait can be modeled, domain-randomized and trained at enormous scale in simulation, then transferred to hardware with a manageable reality gap. Manipulation transfers poorly. Friction, compliance, deformable materials, occluded grasps and slip are precisely the phenomena physics engines approximate worst, so a large share of the training data has to come from the physical world — teleoperation, demonstration or deployed operation — at a cost per hour that does not fall the way GPU hours do.
That is why capital flows toward data infrastructure and embodied-intelligence models rather than actuators. Money buys compute and funds data collection. It does not buy generalization on a schedule.
For an operations leader the practical test is narrow and repeatable. Does performance hold when part presentation changes, when the bin is disordered, when lighting shifts, when a fixture is a quarter inch out of position? Every pilot is asking that question, which is why pilots take as long as they do — and why the companies selling capable platforms to technically sophisticated buyers can ship while the companies selling finished autonomy cannot.
Filter two: unit economics, measured at the cell rather than the robot
Acquisition price is the least interesting number in the deployment decision. The equation that governs is fully loaded cost per task-hour, and the robot’s price is one term in it.
The others: integration and commissioning, fixturing and work cell modification, safety validation, charging or battery-swap infrastructure and the floor space it consumes, uptime, mean time to repair, spare parts and service response time, and — the term most often omitted from public analysis — the supervision ratio. A platform requiring one technician per two units is a fundamentally different business case from one requiring one per twenty, at identical hardware cost.
A robot that works is not the same as a robot that pays. Cycle-time parity with a human operator is necessary and nowhere near sufficient; the machine has to hold that cycle time across a shift and across a quarter, without an engineer resident on the floor.
Filter three: the standards gap
This filter is the least visible in funding announcements and the most likely to govern the timing of scaled deployment.
The governing industrial framework — ISO 10218-1:2025 and ISO 10218-2:2025, the latter having absorbed the collaborative-application requirements formerly held in ISO/TS 15066 — rests on an assumption that does not hold for this machine class: that a robot is statically stable. Remove power from a conventional industrial arm and it stops where it is. Remove power from a bipedal humanoid and it falls. Fall direction, distance and energy depend on velocity, payload and floor condition, and the existing framework does not address any of it.
ISO/TC 299 has a project under development to close that gap. ISO/CD 25785-1 covers safety requirements for industrial mobile robots with actively controlled stability — legged, bipedal, quadrupedal or wheeled self-balancing platforms. It remains a committee draft rather than a published standard, which means integrators today assemble a compliance case from ISO 12100, ISO 10218 where it applies, IEC 61508 or IEC 62061 for system-level functional safety and ISO 13849-1 for safety-related control functions, rather than working from a single Type C standard written for the machine in front of them.
Note who is writing it. The leadership of the U.S. delegation to ISO 25785-1 has included Kevin Reese of Agility Robotics as project leader and Federico Vicentini of Boston Dynamics, alongside Carole Franklin of the Association for Advancing Automation. The companies with the deepest operational records are also the companies drafting the rules that will govern deployment. That is not a coincidence, and it is not something a larger round can shortcut. Standards participation is earned with operational hours.
Who Crosses Over Next
The forward question is which companies move from pilot into general availability. Five criteria separate them, all technical rather than financial, and all observable from outside.
- Repeat orders from the same site. A second purchase order from a facility that has already run a pilot is the highest-quality signal available, because that customer has seen the failure modes and bought anyway.
- Task breadth per deployment. A platform performing three distinct tasks at one site is closer to general-purpose economics than one performing a single task at ten sites. Breadth is what justifies the humanoid form factor over a purpose-built machine.
- A declining supervision ratio, disclosed. Almost nobody publishes it. The companies that begin to are the ones confident the trend is in their favor.
- Serviceability engineering. Field-replaceable actuators, hot-swappable batteries, documented mean time to repair, spare-parts logistics. Unglamorous, and it decides whether a fleet scales past a dozen units.
- Standards engagement. Active participation in ISO/TC 299 or A3 committee work signals a company planning for certified, permanent deployment rather than an indefinite sequence of pilots.
Watch the purchase orders, not the press releases.
What to Watch Over the Next Twelve Months
The AGLT close. Agility’s merger is expected to close in 2026 pending shareholder approval and regulatory clearance. The first US pure-play listing creates the category’s reference valuation and, more usefully, its first quarterly disclosure of humanoid unit economics.
Unitree as the public comp. Post-listing trading gives the industry its first market-priced read on a lean, shipping-first humanoid business, against which every private valuation in the category will now be measured.
LimX’s Hong Kong listing. A second shipping-first company reaching public markets would establish the pattern rather than the exception.
The first order button in the West. No US or European company currently sells a humanoid to general buyers. Whichever does it first — and at what price, with what supervision requirement — sets the benchmark everyone else is measured against.
ISO 25785-1 progression. When the dynamically-stable-robot standard advances toward publication, deployment planning changes for every integrator in the category. It is the least-covered item on this list and arguably the most consequential.
Funding measures conviction. Deployment measures progress. The $11.7 billion already committed guarantees the race continues. The engineering decides who finishes it — and right now, the quadrant where funding and readiness meet is empty. The companies worth watching are the ones closing the distance, and that distance is measured in operational hours, supervision ratios and published standards, not in valuations.
The dataset behind this analysis is maintained in The Robotic Life’s Humanoid Robotics Funding Tracker, updated monthly with primary-source verification, and the Companies index, which publishes deployment stage for every humanoid platform tracked.
Lars Talbert is the founder of The Robotic Life, a data platform tracking the humanoid robotics economy: 38 robot platforms, the companies building them, and the capital funding them. He brings more than 30 years of market experience and more than 20 years in technology sales to the analysis. His work pairs verified specification data with funding intelligence at theroboticlife.com.
Sources
The Robotic Life, Humanoid Robotics Funding Tracker (sector total, per-company capital raised): https://theroboticlife.com/robotics-funding-tracker/
The Robotic Life, Companies index (humanoid platform deployment stage, stage definitions): https://theroboticlife.com/robot-companies/
Figure AI, Series C announcement, September 16, 2025: https://www.figure.ai/news/series-c
Agility Robotics and Churchill Capital Corp XI, joint merger announcement, June 24, 2026: https://www.agilityrobotics.com/content/agility-robotics-to-go-public-through-merger-with-churchill-capital-corp-xi
SEC filing, joint press release exhibit (Churchill Capital Corp XI), deal terms of record: https://www.sec.gov/Archives/edgar/data/2074973/000121390026071290/ea029548401ex99-1.htm
NEURA Robotics, Series C announcement (up to $1.4 billion, milestone-contingent): https://neura-robotics.com/record-series-c/
Hyundai Motor Group, completion of Boston Dynamics acquisition, June 21, 2021: https://www.prnewswire.com/news-releases/hyundai-motor-group-completes-acquisition-of-boston-dynamics-from-softbank-301316206.html
ISO/CD 25785-1, Robotics — Safety requirements for dynamically stable industrial mobile robots, ISO/TC 299 (under development): https://www.iso.org/standard/91469.html
Tech Briefs, "Safety in Motion: Setting the Standard for Humanoid Robots" (ISO 25785-1 US delegation leadership): https://www.techbriefs.com/component/content/article/53111-safety-in-motion-setting-the-standard-for-humanoid-robots
The content & opinions in this article are the author’s and do not necessarily represent the views of RoboticsTomorrow
Featured Product
