Humanoid: where's the money now?

What are the main company categories in humanoid robotics?

Humanoid robotics is not one market.

The money is currently splitting between companies building complete robots, companies building the “brain” that lets robots generalize, companies manufacturing bodies at scale, and companies selling the pieces and data that make humanoids deployable.

One rule matters more than before: many of the newest deals go to cross-robotics suppliers. Mecka, General Intuition, RealSense and PickNik sell to robot arms, wheeled machines and humanoids alike, so they belong to the enabling stack, not to a pure humanoid total.

Before we try to understand where the money is going, let’s understand what the categories are.

Category	What it means	Example companies
Full-stack industrial humanoid OEMs	Companies building complete humanoid robots for factories, warehouses, logistics and industrial work. They own hardware, software and deployment.	Figure AI, Apptronik, Agility Robotics, Neura Robotics, Boston Dynamics, Humanoid
China-scale humanoid OEMs	Chinese companies pushing lower-cost humanoid bodies, faster iteration cycles and higher unit volumes.	Unitree, AgiBot, UBTech, XPENG, EngineAI, Fourier Intelligence
Robot foundation-model companies	Software-first companies building general-purpose robot brains, policies and physical AI models that can work across robot bodies.	Skild AI, Physical Intelligence, General Intuition, Nvidia Isaac GR00T, Google DeepMind Gemini Robotics
Automotive and factory deployment platforms	Automakers and industrial operators using factories as the first serious proving ground for humanoids.	BMW, Mercedes-Benz, Hyundai, Tesla, Toyota, Foxconn
Warehouse and logistics humanoids	Robots optimized for repetitive warehouse work such as tote movement, unloading, sorting and material handling.	Agility Robotics, Apptronik, Reflex Robotics, GXO partners, Amazon pilots
Dexterous hands and manipulation hardware	Companies building hands, grippers, tactile systems and end-effectors that let humanoids do useful work beyond walking.	Sharpa, Shadow Robot, Sanctuary AI, Tangent Robotics, AIDIN Robotics, Tesla Optimus hand stack
Actuators, joints and motion components	Suppliers of joints, motors, reducers, sensors, batteries and power systems that determine cost, torque and reliability.	Schaeffler, Bosch, maxon, RobStride, Tihu, PhyArc
Training, simulation and robot-data infrastructure	Tooling for teleoperation, imitation learning, synthetic environments, robot data capture and fleet learning.	Nvidia Isaac/Cosmos/GR00T, Mecka, World Labs, Neura “gyms,” 1X world models
Home and service humanoids	Humanoids positioned for homes, hospitality, facilities, entertainment or general consumer assistance.	1X, Tesla Optimus, Figure, Fourier, EngineAI, UBTech
Healthcare and assisted-living humanoids	Robots aimed at rehabilitation, eldercare, mobility support, clinical assistance or care settings.	Fourier Intelligence, Devanthro, Neura Robotics, UBTech, Sanctuary AI

Is money flowing into full-stack industrial humanoid OEMs right now?

Yes, full-stack industrial humanoid OEMs are still one of the clearest money zones in humanoid robotics.

The money is chasing evidence, not just demos. The strongest rounds went to companies that combine a full robot body, industrial use cases, strategic investors and some proof that robots can survive real work.

Figure is still the loudest example. Its September 2025 Series C reportedly brought more than $1 billion at a $39 billion post-money valuation, which is extreme for a company founded in 2022. Figure had disclosed that its Figure 02 robots at BMW Spartanburg loaded more than 90,000 parts across 1,250+ runtime hours and contributed to 30,000+ BMW X3 vehicles. In September 2026, Figure signed an initial $3.5 billion compute commitment with Nscale. That is spending on AI compute, not capital raised, but it shows how much Figure plans to put into its models.

The valuation jump also tells us something. Figure raised a $675 million Series B in February 2024 at around $2.6 billion, then moved to a reported $39 billion post-money valuation in 2025. That is roughly a 15x jump in less than two years. In robotics, that kind of step-up usually requires investors to believe the company is building a data and deployment platform, not just selling robots.

Apptronik shows the same strength with a different flavor. In February 2026, it announced a $520 million Series A extension, bringing its Series A to more than $935 million. Repeat investors included Google, Mercedes-Benz, B Capital and PEAK6, while new investors included AT&T Ventures, John Deere and Qatar Investment Authority. That mix of AI, automotive, telecom, agriculture and sovereign capital suggests Apollo is being priced as a platform that could travel across industrial sites.

Neura gives the category a large European proof point. In June 2026, the German company secured up to $1.4 billion at roughly a $7 billion valuation, with Amazon, Nvidia, Qualcomm, Bosch, Schaeffler and the European Investment Bank among the backers. That is a full-stack cap table: component makers, compute suppliers, a logistics giant and public industrial-policy money.

The UK’s Humanoid added a second European name in July 2026. It raised a $152 million Series A at a $1.35 billion post-money valuation, taking its company-reported total financing to $270 million. Its Schaeffler agreement covers thousands of robots, but completed deliveries of that fleet have not been disclosed.

A new type of OEM-adjacent deal also appeared. In July 2026, Walden Robotics launched with a $300 million seed round at a stated $1.1 billion valuation, backed by Toyota among others. Walden builds general-purpose industrial robots, and it has not said that all of its hardware is humanoid, so that money should not be counted as pure humanoid funding.

So, full-stack industrial humanoid OEMs are definitely where money is flowing today. But the category keeps separating: companies with factories, pilots, strategic investors and deployment data get funded; companies with only impressive videos look weaker every quarter.

Is money flowing into China's humanoid makers like Unitree and XPENG?

Yes, China-scale humanoid OEMs are one of the most important money zones, and in 2026 the money moved into both public markets and large private rounds.

The signal is not only VC funding. In China, money shows up through IPOs, production targets, procurement orders and volume economics.

Unitree is the strongest signal. It listed on Shanghai’s STAR Market on 19 August 2026, after an IPO process in which it reportedly sought about ¥4.2 billion, or roughly $610 million. Prospectus coverage pointed to 2025 revenue of about ¥1.71 billion, up 335%, and adjusted net profit around ¥600 million. Its first quarterly report as a listed company, in October 2026, showed Q3 revenue of RMB665.0 million, up 69.93%, while attributable net profit fell 17.37% to RMB112.7 million as R&D spending rose. A profitable manufacturer in a cash-burning field is rare; we compare which business models are winning in our report on humanoid robots.

The shipment signal is also unusually concrete. Unitree reportedly sold more than 5,500 humanoid robots in 2025, and its CEO has discussed shipping up to 20,000 in 2026, which remains a target. Counterpoint Research estimated in August 2026 that more than 22,000 humanoids shipped worldwide in the first half of 2026, with about 9,700 from AgiBot and over 7,000 from Unitree. Even if many units go to research, education or entertainment, those volumes create supplier pressure and cut component costs.

XPENG is the new private-capital signal. In August 2026, its robotics business said it had entered share-purchase agreements for more than $900 million at a post-money valuation above $6.3 billion. Full closing of those agreements has not been independently confirmed. XPENG expects its IRON humanoid to start mass production by the end of 2026 and reach customers in 2027; both are forecasts.

UBTech adds a revenue signal. In August 2026, it reported RMB590.3 million of first-half revenue from full-size humanoid products and services, against RMB38.2 million a year earlier, while the group still posted a loss of RMB338.8 million. That shows Chinese humanoid demand coming through procurement, not just investor presentations.

So China is not only producing viral robot videos; it is building the first large-volume humanoid supply base, and now it has public and private capital behind it. The open question is how much of that volume converts into productive labor. Counterpoint put manufacturing at only 13% of first-half shipments.

Is money flowing into robot foundation models right now?

Yes, robot foundation models are still the most leveraged money zone in humanoid robotics, and the first revenue figures are now appearing.

Investors are not only funding the robot body. They are funding the model layer that could control many different bodies.

Skild AI is the clearest signal. In January 2026, it announced close to $1.4 billion in Series C funding led by SoftBank at a valuation above $14 billion. In September 2026, Skild said it had crossed $100 million in annual recurring revenue ten months after its first commercial deployment, with more than 60 paying customers and $50 million of revenue recognized since launch. These are company-stated, unaudited figures, and its customers use many robot types, so this is not humanoid-only revenue.

That still matters. It is one of the first signs that a robot-brain company can sell software at scale, not only raise money on promise.

Physical Intelligence shows how fast the pricing moves. The company raised $400 million in 2024 at a reported $2.8 billion valuation. In March 2026, reports said it was discussing about $1 billion at a valuation above $11 billion. In October 2026, Bloomberg reported talks to raise at least $1 billion at about $20 billion. Those are talks, not a closed round, and final terms have not been disclosed.

The field is widening. In September 2026, General Intuition raised $220 million at a stated $6.2 billion valuation for action models trained heavily on gaming data. PitchBook-cited coverage had already counted more than $2.2 billion going into robot foundation-model startups in 2025. Nvidia’s Isaac GR00T work points the same way: the industry is organizing around reusable robot intelligence, simulation and reference hardware.

At the end of the day, robot foundation models are where investors make the most leveraged bet. A full humanoid OEM has to build, support and deploy robots one fleet at a time; a strong model could sit across many bodies, customers and workflows. The tension is that this money is not really humanoid money: these companies win whether the winning body has legs, wheels or a fixed base.

Is money flowing into automotive and factory deployment platforms right now?

Yes, money is flowing into automotive and factory deployment platforms. Much of it sits inside strategic relationships rather than classic startup rounds.

The best signals here are factory pilots, repeat corporate investors, industrial offtake agreements and internal production commitments.

BMW is the most useful case because it has become a real test bed, not just a press-release customer. Figure’s BMW Spartanburg deployment produced the strongest Western production metrics so far: more than 90,000 parts loaded, 1,250+ runtime hours and 30,000+ vehicles touched. In 2026, BMW also started or planned humanoid work in Leipzig with Hexagon’s AEON robot for battery and component manufacturing. BMW is not treating humanoids as a one-vendor curiosity.

Mercedes-Benz appears both as a partner and an investor. It returned as a repeat investor in Apptronik’s 2026 Series A extension. A carmaker investing again is stronger than a pilot announcement because it wants economic exposure to the supplier’s roadmap.

Toyota became a clearer money signal in 2026. It backed Walden Robotics’ $300 million seed round in July, and Toyota Ventures co-led Tangent Robotics’ pre-seed in September. Toyota Motor Manufacturing Canada had also signed a commercial agreement with Agility in February 2026 after its pilot.

Schaeffler and Bosch give a component-and-manufacturing angle. Schaeffler and Neura announced a partnership in November 2025 covering key humanoid components, robot offtake and integration into Schaeffler’s production network. Both companies are also industrial partners of Humanoid, which raised its Series A in July 2026.

Hyundai and Boston Dynamics add the corporate-balance-sheet angle. Boston Dynamics said Hyundai planned to purchase tens of thousands of robots, and in September 2026 it opened a robotics application center for factory-training tasks. Tesla said in October 2026 that it had started early production-quality Optimus builds at Fremont; Elon Musk’s 20,000 robots per week by the end of 2027 is a forecast.

Automotive factories remain where humanoid robotics looks most financially legible. The tasks are repetitive, the labor pain is measurable, and the buyer can compare a robot against line downtime, worker shortages and ergonomics. The weak point is that “thousands of robots” agreements still do not tell us how many units are already paid for and running.

So factory deployment is less a separate category than a filter: if a humanoid company cannot find industrial work here, its funding story becomes much harder.

Is money flowing into warehouse and logistics humanoids right now?

Yes, warehouse and logistics humanoids attract money because the use case is specific enough to measure.

This category gets funded when robots can move totes, unload carts, sort items or handle repetitive material flows.

Agility Robotics is the anchor. GXO says Digit has moved more than 100,000 totes at its Georgia site, one of the few humanoid metrics that reads like an operating KPI rather than a lab benchmark. In October 2026, GXO agreed to expand Digit to two more US sites, in Statesville, North Carolina, and Lebanon, Tennessee, with operations planned for early 2027. A repeat order from an existing customer is a stronger signal than a demo.

The money side has moved to public markets. Agility reportedly raised a $400 million Series C in 2025 at about a $2.1 billion valuation, then announced in June 2026 a plan to go public by merging with Churchill Capital Corp XI, at a proposed value of about $2.5 billion. The SEC declared the registration effective in October, but the merger had not closed as of late October. Agility also cites more than $300 million in conditional, multi-year Digit 5 orders, which are potential contract value, while its filings show only $0.174 million of sales in the first half of 2026.

That gap is the tension. Logistics buyers are signing, but recognized revenue is still tiny, and Digit 5 itself is due in 2027.

GXO’s behavior is also useful because it has tested more than one humanoid vendor, including Agility, Apptronik and Reflex Robotics. Buyer demand is not tied to one startup’s marketing. Large logistics operators are scanning for humanoids that fit existing facilities without rebuilding the warehouse around automation. Walden’s July 2026 funding also targets flexible robots for production and logistics.

Amazon is another indirect signal. Its earlier work with Digit and its broader robotics operations show how humanoids might enter warehouses: not as replacements for all automation, but as flexible labor where fixed automation is too expensive or too rigid.

All things considered, logistics looks like one of the most credible early revenue pools. The money is not as spectacular as in foundation models, but the proof is more grounded because the task can be counted.

Is money flowing into dexterous hands and manipulation hardware right now?

Yes, and more visibly than before: dexterous hands and tactile sensing now attract direct rounds and strategic stakes, though the amounts are still small next to OEM rounds.

The reason this category matters is simple: walking makes humanoids impressive, but manipulation makes them economically useful.

The most important strategic signal came from Nvidia’s Isaac GR00T reference humanoid in May 2026. Nvidia combined a Unitree H2 Plus body, Sharpa five-fingered hands, Jetson Thor compute and Isaac GR00T models into an open reference design. Nvidia did not present dexterity as an add-on; it made hands part of the standard architecture. Hands entering a standard reference design show where attention is moving; we look at the biggest opportunities in our humanoid robots market report.

Direct money followed. In September 2026, HD Hyundai Robotics invested KRW13 billion in AIDIN Robotics, a maker of tactile and force sensors, as a minority stake. The same month, Tangent Robotics raised a $4.5 million pre-seed led by Fly Ventures and Toyota Ventures for finger sensing and fine motor learning. Qualcomm also agreed in September to buy PickNik, a manipulation-software company; the price was not disclosed and the deal had not closed as of late October.

The research signal keeps moving too. A 2026 Ruka-v2 paper introduced an open-source humanoid hand with wrist mobility and finger abduction, showing a 51.3% cut in completion time and a 21.2% rise in success rate versus the earlier Ruka hand. A separate UniDex paper presented a 50,000-trajectory dataset across eight dexterous hands for cross-hand transfer. The bottleneck has moved from “can the robot stand?” to “can the robot use its hands?”

The non-obvious point is that hand and touch suppliers benefit from humanoid fragmentation. If there are many robot bodies, a strong hand or sensing supplier can sell into many OEMs, which is a better position than being locked inside one robot company.

So dexterous manipulation is turning from a technical theme into a funded one. The rounds are small, but they target the bottleneck that decides whether humanoids can do factory work.

Is money flowing into actuators, joints and motion components right now?

Yes, but mostly through strategic positioning, supplier partnerships and manufacturing agreements. No large stand-alone actuator or joint round was disclosed between July and October 2026.

This is the layer where humanoid robotics becomes an industrial supply-chain problem.

Schaeffler is the clearest example. In late 2025, it announced a partnership with Neura to develop and supply key humanoid components, with an offtake agreement and integration into Schaeffler’s production network. Neura’s June 2026 fundraise then included Schaeffler and Bosch among investors. First strategic partnership, then capital: component suppliers are trying to secure positions inside future robot bills of materials.

Bosch is also moving from components into manufacturing. In May 2026, Humanoid announced that Bosch would manufacture its HMND 01 robots for Europe while Schaeffler would supply components. This is a very different model from Figure’s vertical integration: humanoid brands can use mature industrial suppliers to cut manufacturing risk.

China’s volume makes this category even more important. With Counterpoint estimating more than 22,000 humanoid shipments in the first half of 2026, demand for joints, motors, reducers, batteries and sensors is no longer theoretical. Component suppliers benefit from the volume race even if the winning robot brand changes. HD Hyundai’s stake in AIDIN is adjacent, but it is a sensing deal, not proof of an actuator funding boom.

This is where the economics get brutal. A humanoid can have strong AI, but if actuators are too expensive, too hot, too noisy or too fragile, the robot will not scale. The component layer is one of the best “quiet money” areas in humanoids: barely visible in funding headlines, but central to whether the category reaches mass deployment.

Why are robot-data startups like Mecka raising big rounds now?

Because robot data has become the scarce input, and it is now one of the clearest new money clusters in the stack.

The market has learned that manufacturing more bodies is not enough; the scarce asset is reusable physical-world learning.

Mecka is the clearest signal. In October 2026, it raised a $60 million Series B led by Sequoia, with Nvidia, M12, Qualcomm Ventures and Samsung joining, to supply robot-learning data and deployment tools. Mecka says it passed a $100 million revenue run rate in June 2026 and projects $300 million by year-end. The first is a company-stated run rate, the second a target.

Larger deals sit just next door. General Intuition raised $220 million in September 2026 for action models built on interaction data. In the same month, AMD agreed to buy World Labs, a spatial world-model company, for about $8.2 billion in stock, and Cognex agreed to buy the RealSense depth-camera business for $500 million in cash. Both are pending acquisitions, not venture rounds, and neither is humanoid-only.

Real deployments feed the same loop. Figure’s 1,250+ runtime hours at BMW also created training data: every failed grasp and successful repetition can feed the next model. Neura’s June 2026 funding supports the Neuraverse platform and robot “gyms” where humans in motion-capture suits teach robots tasks.

Nvidia makes the same bet from the infrastructure side. Isaac GR00T, Cosmos, Jetson Thor and reference designs create a stack for simulation, training and deployment that Nvidia can monetize whichever robot body wins. 1X adds world models and video-based learning for NEO, because teleoperation does not scale if every new task needs a human at the controls.

So the money here is now visible, not hidden. The catch is that it is shared with every other kind of robot, so it should not be added to humanoid totals.

Is money flowing into home and service humanoids right now?

Some money is flowing into home and service humanoids, but this category is much less proven than industrial humanoids, and no large dedicated home-humanoid round was disclosed between July and October 2026.

The home story is attractive because the market could be enormous, yet the evidence is still closer to preorder, partnership and investor belief than scaled consumer use.

1X is the main signal. In 2025, reports said the OpenAI- and EQT-backed company was seeking up to $1 billion at a $10 billion valuation to bring NEO into homes. That is a large number for a consumer humanoid thesis when the category has not yet shown mass-market retention, safety or autonomy. 1X takes orders online, but delivered units and home revenue have not been disclosed.

The most telling 1X signal is that its big deployment channel is not the home. In late 2025, 1X announced a partnership to make up to 10,000 NEO robots available to EQT’s portfolio companies between 2026 and 2030, with a focus on manufacturing, warehousing and logistics. Even a “home” robot may need industrial sites first.

The pricing signal is still worth watching. 1X has discussed a roughly $20,000 purchase price or a $500/month subscription. The missing proof is whether autonomy can justify that cost without constant human supervision.

Figure and Tesla add attention but not customer evidence. In September 2026, Figure showed its Helix 2.5 model working in 30 homes it had never seen, a research test rather than a paid rollout. Tesla started early production-quality Optimus builds in October 2026, but outside sales have not been disclosed.

At the end of the day, home and service humanoids are fundable because the upside is huge. But the most credible money treats home as the long-term prize and commercial deployment as the bridge.

Is money flowing into healthcare and assisted-living humanoids right now?

Not strongly yet. Healthcare and assisted-living humanoids have a clear long-term need, but the capital signals are much weaker than in industrial humanoids, robot foundation models or China-scale manufacturing.

Fourier Intelligence is the best example because it has rehabilitation roots rather than a generic “robot for everything” pitch. Tracker data puts Fourier at more than $200 million raised, an $800 million-plus reported valuation and 500+ units shipped in 2025. Chinese coverage also reported a nearly 800 million-yuan Series E in early 2025, with investors including Guoxin Investment, Pudong Venture Capital and Prosperity7 Ventures. That is meaningful, but not in the same league as the largest rounds. Healthcare rounds trail Figure, Skild, Neura and Apptronik by a wide margin; we follow where the biggest rounds go in our humanoid robots report.

The newest money is public and small. The European Commission signed a €300,000 EIC grant in July 2026 for Devanthro’s Robody, an elderly home-care humanoid project that started in September 2026. It funds pre-commercial validation, not proven clinical deployment.

There is also a small clinical-research signal. A 2026 paper described a teleoperated Unitree G1 providing visualization during a cadaveric endoscopic procedure. The study framed this as proof of form-factor feasibility, not autonomy. Healthcare humanoids can show intriguing demonstrations, but the path to regulated clinical use is much longer.

Neura sometimes shows up in healthcare discussions, and its order backlog reportedly exceeds $1 billion across sectors. But its financing is better understood as physical AI and industrial capital, not a healthcare boom.

So healthcare humanoids look like a future category rather than a current money center.

So, where is the money in humanoid robotics now?

The money in humanoid robotics is still concentrated in three places: robot foundation models, full-stack industrial humanoid OEMs and China-scale humanoid manufacturers.

Those categories have the strongest mix of valuation jumps, large rounds, strategic investors, production data and deployment proof. Foundation models keep first place because Skild now shows revenue and Physical Intelligence is reportedly pricing near $20 billion, but much of that money serves all robots, not just humanoids.

The biggest move is in robot data. Mecka, General Intuition and the pending World Labs and RealSense acquisitions push training and data infrastructure ahead of factory platforms in our ranking. Dexterous hands also move up, because they now attract direct rounds.

Home and healthcare remain exciting, but the evidence today is thinner. And no total by category can be added up honestly: signed-but-unclosed rounds, IPO proceeds, acquisition prices and cross-robotics deals all need to stay separate.

Rank	Category	Signals proving the money is here
1	Robot foundation models	Skild raised close to $1.4B at $14B+ and says it passed $100M ARR (company-stated); Physical Intelligence reportedly in talks for $1B+ at about $20B (not closed); General Intuition raised $220M at $6.2B.
2	Full-stack industrial humanoid OEMs	Figure reached a reported $39B post-money valuation; Apptronik reached $935M+ in Series A funding; Neura secured up to $1.4B at roughly $7B; Humanoid raised $152M at $1.35B.
3	China-scale humanoid OEMs	Unitree listed in Shanghai in August 2026 and reported Q3 revenue up 69.93%; XPENG robotics signed agreements for $900M+ at $6.3B+; UBTech reported RMB590.3M of H1 humanoid revenue.
4	Training, simulation and robot-data infrastructure	Mecka raised a $60M Series B; AMD agreed to buy World Labs for about $8.2B in stock (pending); Nvidia is building the GR00T/Cosmos/Jetson stack; most deals serve all robots.
5	Automotive and factory deployment platforms	BMW, Mercedes, Toyota, Schaeffler, Bosch and Hyundai are moving through pilots, repeat investments, offtake agreements and manufacturing roles; Toyota backed Walden’s $300M seed.
6	Warehouse and logistics humanoids	GXO agreed to two more Agility sites for early 2027; Agility’s SPAC merger at a proposed ~$2.5B had not closed; $300M+ in multi-year orders against $0.174M of H1 2026 sales.
7	Dexterous hands and manipulation hardware	HD Hyundai invested KRW13B in AIDIN’s tactile sensing; Tangent raised $4.5M; Qualcomm agreed to buy PickNik (pending); Nvidia made Sharpa hands part of its reference humanoid.
8	Actuators, joints and motion components	Schaeffler and Bosch hold partnership, investment and manufacturing roles; no large stand-alone actuator round disclosed from July to October 2026.
9	Home and service humanoids	1X reportedly sought $1B at a $10B valuation and signed a 10,000-robot EQT channel; Figure’s 30-home test is research, not sales; no large dedicated home round disclosed.
10	Healthcare and assisted-living humanoids	Fourier has healthcare roots and meaningful funding; Devanthro won a €300,000 EU grant; early surgical-assistance research is emerging, but capital is not concentrated here.