Inside the $11B Humanoid Robotics Cash Wave

Zach Martin

Welcome to Robot Capital, on KMKT IR Hub Radio. I'm Zach Martin. It's October 8, 2026, and today's job is simple to say and harder to do. We're going to follow the money moving into robots, and then ask what that money is actually buying. We'll cover the venture numbers, the biggest rounds, the industries pulling capital in, the public market tests that are arriving right now, and the commercial evidence that either supports all this spending or doesn't yet. I'll keep the facts and my interpretation in separate boxes, and I'll tell you when two sources disagree, because in this sector they disagree a lot. One housekeeping note. Nothing here is investment advice. Think of this hour as a coach walking you through game film, not a coach telling you which player to bet your savings on. Game film is useful because it shows you patterns, and the pattern this week is a loud one. So let's start with a single number, because everything else hangs off it.

Zach Martin

The number is 11.0 billion dollars. That's how much venture capital has gone into humanoid robotics startups so far in 2026, according to Dealroom, with data running through October 6. Now look at what came before. The full year 2025 set a record at 4.8 billion dollars. Before that, 2024 was about 1.5 billion. In 2023 it was 556 million, and in 2022 it was 553 million. So for two straight years the number barely moved, then it roughly tripled, then it more than doubled again, and we still have almost three months left on the calendar. Here's the analogy I'd use. Imagine a runner who jogs the same easy pace for two years, then suddenly doubles her speed, and then doubles it again. You wouldn't just say she got faster. You'd want to know what changed in her training, who's paying for the shoes, and whether her knees can handle it. That's the posture for this whole episode. Dealroom counts announced venture rounds into venture backed companies carrying its humanoid tag, and it assigns each company by headquarters. Keep that definition in mind, because other trackers count differently, and we'll see the numbers wobble a little between them.

Zach Martin

Let's look at what's inside that 11 billion. Dealroom's list of the largest 2026 rounds starts with Neura Robotics in Germany, a Series C of 1.4 billion dollars. Next is XPeng Robotics in China, a 900 million dollar Series A. Then AI2 Robotics at 735 million, Sharpa at 619 million, Apptronik at 520 million, Rhoda AI at 450 million, and Galbot at 344 million. Now, if you've seen other reports, you might notice small differences. The other tracker I'm leaning on, from Newmarketpitch, puts AI2 Robotics at roughly 739 million, Sharpa at 667 million, and Galbot at about 365 million. Same companies, same period, slightly different totals. That's normal. Trackers disagree on when a round closes, whether an extension counts as part of the original round, and how to convert currencies. My rule is to treat these as ranges. When I say about 700 million for AI2, I mean somewhere between the 735 and 739 that the two sources report. The direction of travel is what matters here, and on direction they agree completely. Hundreds of millions of dollars per round is now ordinary at the top of this market.

Zach Martin

Humanoids are only one slice of a wider robotics world, so let me widen the lens for a moment. The research I'm working from logs 167 global robotics funding rounds in September 2026, totaling approximately 4.86 billion dollars. That's a different dataset from Dealroom's humanoid tag, so please don't try to subtract one from the other. What I want you to hear is the combination. Many transactions, spread across automation broadly, and yet the dollars are tilting hard toward a small number of very large checks. That tilt is the real story of the year, and the cleanest evidence comes from Newmarketpitch's quarterly analysis. Q2 2026 set an all time quarterly record for measured robotics capital at 3.79 billion dollars. That's up 35.6 percent from Q1's 2.79 billion, and roughly triple Q2 2025's 1.17 billion. Sounds like a pure boom, right? But here's the twist. The number of deals fell 28 percent, from 50 down to 36. More money, fewer deals. When that happens, you should always ask the same question. Who is getting the money, and who isn't?

Zach Martin

Here's the answer. Eleven financings of 100 million dollars or more generated roughly 91 percent of all measured capital in Q2. By my own arithmetic, that's somewhere around three and a half billion dollars from eleven deals, out of 36 total. Picture a potluck dinner where thirty six people show up, but eleven of them bring nearly all the food, and the other twenty five split a few casserole dishes. Everybody technically contributed. Nobody would describe the table as evenly stocked. Now, concentration by itself isn't a scandal. Building a humanoid at scale is expensive, and a company that needs factories, GPU clusters, and fleets of test robots needs big checks. But concentration changes the character of a market. It means the headline total can set a record while most of the companies in the sector feel no tailwind at all. If you're a founder reading that 3.79 billion dollar headline and you're trying to raise a modest round, the headline may have almost nothing to do with your actual weather.

Zach Martin

The evidence for that squeeze is right in the data. The median robotics round peaked at 17.9 million dollars in Q4 2025. It slid to 16.3 million in Q1 2026 and to 15.5 million in Q2. So the typical round is getting smaller even as the total gets bigger. Then there's the middle of the market. Rounds between 20 million and 50 million dollars took 31.3 percent of capital in Q3 2025. In Q2 2026 that share was just 3 percent. And the total raised in rounds under 50 million dollars dropped to 261 million in Q2 2026, down from 469 million in the same quarter a year earlier. Think of a barbell. You've got heavy plates at both ends and almost nothing in the handle. For a startup, the middle of the range is where you're supposed to prove the product works before you ask for the big money. If that middle is hollowing out, the path from a good demo to a funded company gets steeper, and the companies that make it through will increasingly be the ones who can jump straight to the heavy end of the bar.

Zach Martin

One more cut of the Q2 data, and it's the one that I think explains the behavior. Follow on financings, meaning money going to companies that had raised before, took 3.74 billion dollars of the 3.79 billion total. First time financings captured only 49 million dollars. That's about 1.3 percent of the capital, even though first time deals were 19.4 percent of the deal count. So roughly one deal in five was a newcomer, and those deals shared a sliver of a percent of the money. That is a fact. Here's my interpretation, and I'll label it as mine. Investors are not spreading seed bets widely and waiting to see what grows. They're stacking more capital onto teams that have already shown they can execute, and they're doing it quickly. We'll meet a company in a few minutes that raised money three times in about six months. If you think of venture investing as gardening, the gardeners right now aren't scattering seeds across the field. They're dumping fertilizer on the three or four plants that already look healthy and hoping those plants grow into trees. That can work. It can also leave the whole garden dependent on very few plants.

Zach Martin

So we have a picture. Record dollars, fewer deals, a shrinking middle, and a heavy tilt toward follow on checks. The next question is the obvious one. Who exactly is on the receiving end of the biggest checks, and what are they promising to do with the money? Let's go name by name, because the details of each round tell you more than the totals do. We'll start with the company that has become the reference point for the whole sector, and then move through Europe, Texas, Michigan style factory floors, and China. Along the way I'll flag where the money is committed, where it's only announced, and where a label like Series A doesn't mean what you'd expect it to mean.

Zach Martin

Start with Figure AI. According to the funding history in the Newmarketpitch ranking, Figure raised 70 million dollars in its Series A, then a 675 million dollar Series B in 2024. Its Series C then exceeded 1 billion dollars in committed capital at a 39 billion dollar post money valuation, and that figure comes from Figure's own announcement. Investors in that round included Parkway Venture Capital, Brookfield, NVIDIA, Macquarie Capital, Intel Capital, Salesforce, and Qualcomm Ventures, among others. Cumulative funding depends on who's counting. Newmarketpitch says more than 1.7 billion dollars. Dealroom says 1.9 billion raised since 2020. Same company, two trackers, a couple hundred million apart. What matters is the shape. That Series C alone was roughly fifteen times larger than the Series A. Figure said the money would go to Helix, its AI system, to manufacturing through its BotQ facility, to GPU infrastructure, and to much larger real world data collection programs. Notice what's not on that list. It's not just engineers building a better prototype. It's factories, chips, and data. This is a company spending like an infrastructure business.

Zach Martin

In September 2026, Figure also partnered with Nscale to deploy dedicated AI training infrastructure on NVIDIA's Vera Rubin platform for its Helix models. Let me translate that. Figure is not only buying robots and building a factory. It is arranging dedicated computing capacity to train the software brain that runs those robots. If you've ever watched a restaurant owner decide whether to rent a commercial kitchen or build their own, you understand the logic. Renting is flexible. Building your own is expensive, but it gives you control over quality and schedule. At Figure's scale, the bet appears to be that control over training capacity is worth the cost. That's a strategic claim, not a proven one. The proof would show up later, in how quickly Helix improves and how reliably it performs on customer sites. For now, it tells us something about what the 39 billion dollar valuation is supposed to cover. It's supposed to cover a vertical stack, from compute to model to machine.

Zach Martin

Now Europe. NEURA Robotics announced a Series C in June 2026 worth up to 1.4 billion dollars, with backing from Tether, Qualcomm, Amazon, NVIDIA, Bosch, Schaeffler, the European Investment Bank, Lingotto Horizon, and others. The company says the capital will support mass production, its Neuraverse software ecosystem, and a worldwide network of what it calls NEURA Gyms, where robots collect real world training data. Dealroom lists the round at 1.4 billion, the biggest check of the year so far, with NEURA raising 1.7 billion since 2020 at a 7.0 billion dollar valuation. But look at the phrase up to. Newmarketpitch makes a careful point here. A number announced as up to 1.4 billion can include capital whose availability depends on the final structure or on milestones. So the cleanest way to say it is this. NEURA has suddenly gained access to vastly more capital than it had before, but the exact amount already in the bank may be lower than the headline. Compare that to a closed round like Figure's. It's the difference between a pre approved credit limit and cash in your checking account. Both matter. They're not the same.

Zach Martin

What NEURA did next is instructive. In the third quarter, it acquired Bosch Rexroth's ACTIVE Shuttle platform and ADLATUS Robotics. A company that has just raised a mountain of capital goes shopping. That tells you a bit about the strategy. It's not only building a humanoid from scratch. It's assembling a portfolio of industrial robotics capabilities and customers, and the money lets it do that quickly. Also notice the investor list includes Bosch and Schaeffler, which are industrial companies, not just financial firms. We'll come back to why that matters. For Europe, the broader picture is significant. Dealroom shows Europe at 15 percent of 2026 humanoid venture dollars, about 1.6 billion. Before this round, the biggest private rounds in the sector were overwhelmingly associated with American and Chinese companies. NEURA put a European name into the same capital tier, and that's why a lot of the continent's investors and policymakers will be watching how this one turns out.

Zach Martin

Now to Austin, and a good lesson in why labels can mislead. Apptronik announced a 350 million dollar Series A in early 2025. That grew to 403 million after more investors joined. Then in 2026 it added another 520 million dollars, labeled a Series A extension, which brought the full Series A to more than 935 million dollars and total capital raised to nearly 1 billion, according to the company. Dealroom values the company at 5.5 billion. Investors include Google, B Capital, Mercedes Benz, and PEAK6 returning, with AT&T Ventures, John Deere, and Qatar Investment Authority joining the extension. A Series A is traditionally the first big institutional round. Here it has swollen to nearly a billion dollars. The label tells you the company's formal stage. The dollars tell you its actual scale. Apptronik also has commercial relationships that overlap with its financing story. Mercedes Benz has been testing its Apollo robot in manufacturing. Google DeepMind is working with it on robotics AI. Jabil is both a manufacturing partner and a deployment partner. And it has set up its Robot Park training grounds in Austin alongside Google DeepMind. Money, customers, and technology partners are often the same names on one cap table.

Zach Martin

Next, the clearest example of serial follow on dominance, which is Mind Robotics in the United States. It spun out with Rivian manufacturing data behind it. It raised a 115 million dollar seed round in November 2025. Then a 500 million dollar Series A in March 2026, led by Accel and Andreessen Horowitz. Then another 400 million dollars in May 2026, led by Kleiner Perkins. Add those up and you get roughly 1.015 billion dollars in about six months, to build AI driven factory manipulation systems that are tested directly on live Rivian vehicle lines. A similar pattern shows up at Bedrock Robotics, which raised a 67.3 million dollar Series A in July 2025 and then a 270 million dollar Series B in February 2026, backed by CapitalG and Valor Atreides AI Fund, to retrofit construction heavy equipment for autonomous fleets. And there's a report, and I'll stress that it's a report, that Field AI is raising 700 million dollars at a 10 billion dollar valuation in early October 2026. I can't confirm that it has closed. Treat it as a signal about the appetite, not a completed fact.

Zach Martin

Now China, where the center of gravity has moved. Dealroom says China captured 63 percent of global humanoid venture capital in 2026 so far, about 6.9 billion dollars. That's up from 25 percent in 2020. The United States share has contracted to 12 percent, about 1.3 billion. Europe stands at 15 percent and the rest of the world at 10 percent, about 1.1 billion. Notice that Sharpa, which raised between 619 and 667 million dollars in August backed by Alibaba, Tencent, and Meituan at roughly a 3 billion dollar valuation, is listed by Dealroom as Singapore based, so it sits in the rest of the world bucket by headquarters. That's the kind of detail that changes how you read geography charts. Other major Chinese rounds include XPeng Robotics, a 900 million dollar Series A at a 6 billion valuation, AI2 Robotics, with around 918 million dollars raised and a valuation near 2.9 billion, and Galbot, a Series B extension of 344 to 365 million dollars backed by CITIC, Sinopec, and SAIC, with total funding around 999 million. Also RobotEra, which raised more than 200 million dollars, with investors including SF Group and IDG Capital.

Zach Martin

Why China? The usual argument has three parts, and I'll flag that this is an argument, not a settled finding. First, state guidance. One of the market reports I'm using says China's industrial ministry mandates AI integration across ten priority manufacturing sectors. Second, deep corporate backing from giants like Xiaomi, SAIC, SF Group, Alibaba, and Tencent. Third, an end to end supply chain for core hardware, things like frameless torque motors, harmonic reducers, planetary rollers, and tactile arrays. If you can buy the hard parts down the street instead of shipping them across an ocean, you iterate faster. Think of a baker who has the flour mill, the oven maker, and the delivery drivers all in one neighborhood. She can try a new recipe every morning. The depth of the field matters too. Newmarketpitch finds the top ten companies in its ranking account for roughly 12 billion dollars, with 17 companies at or near the 500 million dollar mark, and China has the deepest bench of heavily funded humanoid and embodied AI companies. Meanwhile Dealroom notes that 45 companies founded since January 2025 are already unicorns, mostly AI labs and robotics companies. Its warning is blunt. Backing a one year old unicorn means paying a late stage price for a seed stage company, and that price rests on the team, not on revenue.

Zach Martin

Below the giant rounds, the pattern continues in smaller form. Sunday Robotics raised 165 million dollars in a Series B in March 2026 from Coatue, Bain Capital Ventures, and Fidelity, for a household robot called Memo. A UK company simply named Humanoid raised 152 million in a Series A in July 2026, at a 1.35 billion dollar post money valuation, backed by Schaeffler, Bosch, and Prime Movers Lab. Newmarketpitch's summary is that nine figure humanoid rounds are now appearing repeatedly rather than once or twice a year. Dealroom's list of companies founded in 2026 that are already unicorns even includes Walden Robotics at 1.1 billion dollars. So the picture of the rounds is clear. Now ask the more practical question, the one an operator or a corporate buyer would ask. Where in the real economy is all this capital landing? Humanoid hype tends to make us picture a robot in a kitchen. The actual money is flowing toward a much less glamorous list of workplaces, and that's where we go next.

Zach Martin

Here's a principle I'd write on a whiteboard before looking at any robotics company. Robots go first where the work is measurable. If you can count cycle times, error rates, and labor hours, you can build a business case. If the task is vague, the environment is chaotic, and the success criteria change by the hour, you've got a research project, not a product. That's why automotive and heavy manufacturing are the primary industrial beachhead. The tasks are structured, labor is expensive, and the line is already instrumented. Look at the names. BMW deployed Figure 02 in Spartanburg, South Carolina. Mercedes Benz is testing Apptronik's Apollo. Rivian is operationalizing Mind Robotics on live vehicle lines. And in April 2026, Schaeffler and Hexagon AB announced a partnership to deploy 1,000 AEON humanoid robots across factories by 2032. I want to be careful with that last one. It's a plan, with a date six years out, not a delivered fleet. Plans are useful signals of intent. They're not revenue.

Zach Martin

Logistics is the second big lane, and it's easier to read because warehouses generate dense transaction data. Every pick, every tote, every damaged package can be counted. Agility Robotics' Digit has surpassed 100,000 totes moved at a GXO Logistics facility in Georgia, in live operations. On the funding side, the deals include Dexterity with a 104.9 million dollar Series C, Mytra with a 120 million dollar Series C, Dexory with a 100 million dollar Series C, and SENAD with more than 29 million. The metrics buyers care about here are things like picks per hour and damage rates. That's a different language from the one used in humanoid demo videos, and it's a healthier one. A market report from Persistence Market Research gives a sense of the price tag. It says a mid scale warehouse autonomous mobile robot deployment may require 500,000 dollars to 2 million once you include fleet management, integration, and training expenses. Keep that range in your head. It's going to matter when we talk about why revenue may arrive more slowly than funding.

Zach Martin

Third lane, defense, drones, and security. Mach Industries raised 600 million dollars in September 2026 for autonomous defense infrastructure and drones. ARX Robotics raised about 35.2 million dollars in a Series A for modular unmanned ground combat vehicles. Gecko Robotics raised 125 million in a Series D plus. Seasats raised 20 million in a Series A for autonomous maritime vessels. Persistence Market Research points to active Department of Defense autonomous systems contracts exceeding 4.6 billion dollars awarded across 2023 and 2024, and names defense as one of the demand drivers for physical AI. I'd offer a cautious interpretation. Defense is attractive to capital for the same reason factories are. There's a buyer with a budget and a specific problem. That said, I'd note that the research I have gives limited detail on how these defense rounds convert into contracts, so I won't speculate beyond the headline amounts. What we can say is that the money is no longer confined to humanoid shaped machines. A lot of physical AI is wheels, tracks, hulls, and wings.

Zach Martin

Fourth, construction and heavy equipment. This is the retrofit story. Bedrock Robotics, with its 270 million dollar Series B, is retrofitting construction machinery for autonomous fleets. Gravis Robotics raised 23 million dollars, August Robotics raised 30 million in a Series B, and Raise Robotics raised 7.8 million. Notice the strategy. Instead of inventing a new machine, these companies add intelligence to machines that already exist. That's the same logic BCG describes in its analysis of industrial automation. BCG says that instead of replacing entire production lines, companies can retrain physical AI systems to handle new requirements with minimal hardware changes. It's like adding a smart driver to a truck you already own, rather than buying a whole new truck. Then there's an unusual one. Watney Robotics raised 80 million dollars in September 2026 for autonomous datacenter server rack inspection and physical maintenance. My interpretation, and it's only that, is that the boom in AI computing creates physical buildings full of machines that need physical care, so robots get hired to look after the servers that train other robots. I find that loop a bit funny, and also quite logical.

Zach Martin

Fifth, healthcare and surgery. Distalmotion raised 150 million dollars in a Series D plus. Cornerstone Robotics raised about 200 million. ForSight Robotics raised 125 million in a Series B, Beijing Surgerii raised 100 million, and Zap Surgical raised 11 million for radiosurgery. The demand story is demographic. Persistence Market Research points to aging populations in Japan, Germany, and the United States, and notes that the FDA had cleared more than 1,000 AI and machine learning based medical devices as of 2024. Fortune Business Insights adds a useful caution. It says medical robotics is growing, but clinical validation will constrain unsupervised autonomy. In other words, in a hospital, the machine can assist, but you won't see it freewheeling any time soon. And sixth, agriculture. Ecorobotix raised 105 million dollars in a Series D plus. 4AG Robotics raised about 29 million in a Series B, and Beewise raised 50 million in a Series D plus. Different crops, different environments, same pitch. Labor is scarce, tasks are repeatable, and data accumulates every season.

Zach Martin

What about the home, the thing everybody pictures when they hear humanoid? Home robotics is attracting real money. Sunday Robotics raised 165 million. The Bot Company, which focuses on household chore robots, raised 219 million dollars in October 2025 in a round labeled Series C, bringing it to about 521 million in total. But Newmarketpitch makes an observation I find useful. The billion dollar rounds still cluster around companies that can start with industrial or commercial deployments. The home is the dream customer, and also the hardest environment. Every kitchen is different, every family has different rules, and mistakes happen next to your children and your fragile things. Factories and warehouses are the training gym. Homes are the championship fight. Most capital seems to be funding the gym first. If you're evaluating any company in this space, I'd ask a coach's question. Which environment is it starting in, and is that environment a stepping stone or a detour?

Zach Martin

Now the most important question in this chapter. What do the mega rounds actually pay for? The research gives a consistent answer. The money is increasingly paying for infrastructure rather than just prototypes. Factories, robot fleets, GPU capacity, data collection, training facilities, and supply chain scale now absorb hundreds of millions of dollars before these companies have anything close to mass market revenue. Look at the evidence we've already heard. NEURA's Gyms. Apptronik's Robot Park. Figure's BotQ plant and its Nscale training deal. Mind Robotics testing on live Rivian lines. Think of it as building a restaurant's kitchen, training center, and supply chain before the first dinner service. You can't open without the kitchen. But a kitchen is a cost, not a customer. The danger of a capital intensive build is simple. If the customers show up late, the fixed costs keep running. That's why Fortune Business Insights calls physical AI a capital allocation question, asking which workflows can generate durable productivity gains after integration, safety, maintenance, and workforce costs are included.

Zach Martin

One more macro thread belongs in this chapter, because it explains why industrial buyers are listening. BCG highlights Foxconn's brownfield automation work, where physical AI systems are retrained in simulation to handle component variability. The context is nearshoring. BCG notes that shifting production out of mature hubs like China into destinations such as Mexico, India, and Vietnam typically imposes efficiency losses of 4 percent to 15 percent. Physical AI and adaptive automation are being deployed to help bridge that gap, and to manage structural demographic labor constraints. So here's the buyer's logic, stated plainly. If moving a factory costs you a double digit slice of efficiency, and workers are harder to find, then a flexible machine that adapts without rebuilding the line suddenly looks like insurance. It's an interpretation, but it's grounded in how BCG frames it. And it points to an uncomfortable tension, because the geography of capital and the geography of supply chains are pulling in different directions. China attracts most humanoid venture dollars while companies elsewhere try to diversify production away from it. Capital and industry are not always marching in step.

Zach Martin

Let's talk about who writes the biggest checks, because it changes how you should read every headline. Dealroom reports that the Magnificent 7 took part in rounds raising 308 billion dollars in 2026, which is 45 percent of all venture dollars raised worldwide. But they were part of only 1.3 percent of the deal count. Same pattern we saw in robotics. Huge dollars, small number of deals. And here's the part to keep in perspective. Four rounds are the reason. OpenAI at 122 billion, Anthropic twice, at 65 and 30 billion, and xAI at 20 billion. Together that's 237 billion of the 308. Those are AI model companies, not robot makers. So when you hear that Big Tech is in nearly half of venture dollars, don't picture robots. Picture a handful of enormous AI labs financed in large part by firms that also sell them compute. Dealroom also notes that only four of the seven actually take part in venture rounds. Apple, Meta, and Tesla have joined 13 rounds between them since 2023, while Alphabet has joined 496. The seven are a stock market grouping, not a dealmaking club.

Zach Martin

Within that group, the name that matters most for robotics is NVIDIA. Dealroom says NVIDIA took part in 85 venture rounds in 2026 so far, more than any other of the seven, and backs 13 of the 45 unicorns founded since January 2025, more than any venture firm. It is an investor in Figure, NEURA, and Agility, among others. And it isn't just an investor. In March 2025, NVIDIA launched Isaac GR00T N1, described as the first open, customizable foundation model purpose built for humanoid robots. It supplies simulation tools through Omniverse and Isaac. The research brief adds its Cosmos models and a safety architecture called Halos for Robotics. So think of NVIDIA as a kingmaker in the sense of a toll road. It sells the chips, the software, and the simulation, and it also owns a seat in many of the companies that buy them. Dealroom's reading is that for NVIDIA, backing these companies means a seat with the labs that will buy its chips. That's an interpretation from Dealroom, and it seems fair. The practical takeaway for you is that some of this capital circulates. Money invested in a robotics startup can come back as chip and cloud spending.

Zach Martin

The other tech giant that shapes the field is Google. Google DeepMind introduced Gemini Robotics in March 2025, extending multimodal Gemini capabilities into robotic control. The brief also points to Gemini Robotics 2 and embodied reasoning models, and to a close partnership with Apptronik. Figure introduced its Helix system in February 2025. Fortune Business Insights frames these moves as a shift toward models that interpret environments and generate actions, rather than models that classify text or images. The dominant technical idea is called vision language action, or VLA, a model that links what a robot sees, what it's told, and what it does. Alphabet also runs the largest venture operation of any company in the world, according to Dealroom, across sixteen vehicles, with 89 rounds so far in 2026. So when Apptronik announces Google as an investor and a technology partner, that's a three in one relationship. Investor, research collaborator, and signal to other investors. When the same logo appears in all three places, it's worth asking how much independent validation you're actually seeing.

Zach Martin

That brings us to strategic money in general. Newmarketpitch finds that corporate and strategic participation reached 41.7 percent of Q2 2026 robotics rounds. And the median strategic backed round jumped from 21 million dollars in Q3 2025 to 149 million in Q2 2026. That's a seven fold increase in the typical check in less than a year. Now, strategic investors can be good news. A carmaker or a parts supplier brings a factory to test in, real data, and a path to a first customer. We've seen Bosch and Schaeffler show up in more than one round, and Mercedes Benz, Rivian, and John Deere appear on cap tables. But here's the caution, and it's my interpretation. When the investor is also the customer, the supplier, or the technology partner, the round can overstate independent demand. It may be a real commitment, or it may be an early option the company is paying for with equity. The best way to tell is to look for evidence that someone paid cash for a deployed robot doing real work, which is the question we'll tackle in the last chapter.

Zach Martin

Now the public market test, which is where talk becomes numbers someone has audited. In August 2026, Unitree Robotics completed its IPO on Shanghai's STAR Market, raising about 6.1 billion yuan, roughly 905 million dollars, at 150.8 yuan per share. The listing revealed audited financials. Unitree generated about 1.7 billion yuan in 2025 revenue, up more than four times from the prior year, which is around 240 million dollars. That reflects high physical shipment volumes. Newmarketpitch notes Unitree has much higher physical shipment volume than Figure or Agility, though it also notes the data confidence on Unitree's funding is only partial. The founder, Wang Xingxing, has taken a pragmatic stance, cautioning that full general purpose autonomy remains years away. I find that quote refreshing. The person with the most hardware in the field is telling you not to expect magic soon. And the brief describes Unitree's market value as in the tens of billions, set against that roughly 240 million dollars of revenue. I'd treat that as an approximate characterization and check the live quote yourself, because public valuations move daily.

Zach Martin

Unitree isn't alone as a listed name. UBTECH Robotics has been public in Hong Kong since late 2023, and it has repeatedly gone back to the public market. It completed share placements generating 2.41 billion Hong Kong dollars and 3.06 billion Hong Kong dollars, and most recently about 139 million dollars in post IPO equity in September 2026, bringing its total capital to roughly 1.6 billion dollars. Newmarketpitch's insight is that the funding race is starting to look less like a normal startup leaderboard and more like a capital markets competition. A listed company can raise again and again, if investors keep wanting the stock. That's a different kind of discipline. Private valuations are negotiated in a room with a handful of investors. Public valuations get repriced every morning by strangers. It's like the difference between practicing alone in the gym and playing in front of a crowd. Some athletes thrive on that. Others discover that practice numbers don't survive the crowd. We're about to find out which robotics companies are which.

Zach Martin

The next test is Agility Robotics. Agility held its Analyst and Investor Day on Tuesday, October 6, 2026, ahead of a planned public listing through a SPAC merger with Churchill Capital Corp XI, ticker CCXI. A SPAC, for anyone new to the term, is a company that is already listed and has no operations of its own, which merges with a private company to bring it public. It's a faster route than a traditional IPO, though not necessarily a smoother one. Agility also announced a safety partnership with FORT Robotics, which itself is preparing a public SPAC debut. For context, Agility's Digit has the logistics track record we discussed, 100,000 totes at GXO. Newmarketpitch lists about 578 million dollars raised, including a 400 million dollar Series C with SoftBank, NVIDIA, and Amazon among the investors, and Dealroom puts its valuation at 2.5 billion. I don't have the details of what management presented on Tuesday, so I won't characterize the targets. What I'd watch in the filings are plain things. Number of deployed robots, revenue, how much cash comes in at closing, and how much leaves through redemptions.

Zach Martin

Finally, consolidation, because exits are the other half of the capital cycle. September brought several deals. Cognex acquired Intel's RealSense 3D depth sensing business for about 500 million dollars in cash. Qualcomm acquired PickNik Robotics, an open source robotics framework company. And SoftBank Group acquired full ownership of the Robotics and AI Institute from Hyundai, a deal that is pending CFIUS review, meaning a United States government review of foreign investment. Add NEURA buying parts of Bosch Rexroth's shuttle platform and ADLATUS. Here's my reading, labeled as interpretation. The acquirers are buying the picks and shovels, sensors, software frameworks, and research teams, rather than betting on a single winning robot body. That's what you'd expect in a gold rush. The people selling the tools often get paid whether or not any individual miner strikes gold. So there's your set of exits. A few listings, a few strategic acquisitions, and a lot of private companies waiting for their turn. Now let's ask whether the commercial evidence is strong enough to justify the rush.

Zach Martin

Time for the reality check, and I want to approach it like a coach reviewing the stat sheet instead of the highlight reel. Let's begin with the best measured example we have. Figure 02 at BMW's Spartanburg plant accumulated over 1,250 operating hours, handling more than 90,000 sheet metal parts across ten hour weekday shifts, and contributing to more than 30,000 X3 vehicles. That's real. It's one of the strongest factory data sets in humanoids, according to Newmarketpitch. Digit's 100,000 totes at GXO is another real one. And Unitree's shipments are a third kind of traction. Now notice what these three are. A factory work record, a logistics work record, and a hardware sales record. They are three different kinds of traction, and none of them is the thing the largest valuations imply, which is thousands of general purpose robots earning recurring revenue. 1,250 hours is about fifty two days of round the clock operation. That's meaningful for a pilot. It's small next to a company valued at 39 billion dollars, and I'd say so about any company in this field.

Zach Martin

Newmarketpitch puts the gap in one sentence that I'll quote. The humanoid robotics market is real, but customer deployments are still tiny compared with the production capacity investors are funding. And the next test is whether pilots, pre orders, and research shipments turn into recurring deployments measured in thousands of useful robots rather than dozens. Hold that thought, because it's the question underneath this entire episode. Think of building a stadium. The stadium may be gorgeous and may even be needed. But until the seats are full on a regular schedule, you don't have a business. You have a very expensive building. The capital is building the stadium. The deployments are the crowds. Right now the stadium plans are bigger than the crowds, and that's not a criticism. Every infrastructure build looks like that at the beginning. It's just that the time lag between building and filling is where companies run out of cash, which is why the funding pattern we saw earlier matters so much. When only a few companies can raise the huge checks, only a few can afford the lag.

Zach Martin

So what's the technical bottleneck, the thing that decides how fast the crowds arrive? The widely held view in this research is that walking and bipedal balance are solved baselines. The new frontier is dexterous manipulation, meaning hands, tactile sensing, high speed multi fingered grasping, and end to end vision language action models. Companies like PaXini, Sharpa, and Astribot sit in that fight. Newmarketpitch says investors are putting more weight on hands, tactile sensing, and manipulation models because useful grasping and object handling remain hard to make reliable. Fortune Business Insights goes further and names reliability as the largest restraint. Physical systems encounter contact, obstruction, sensor degradation, changing layouts, network latency, and human activity that benchmarks do not fully capture. And it says collecting interaction data in the physical world is slower than collecting digital training data. Here's the plain English version. A language model learns from the internet, which already exists. A robot learns from the world, which has to be visited one grasp at a time. That's why NEURA Gyms and Apptronik's Robot Park exist. They're data farms.

Zach Martin

Now the cost trap that doesn't make the headlines. The International Federation of Robotics states that programming and system integration can account for 50 percent to 70 percent of the total cost of a robot application. Let that sink in. If a robot costs a certain amount, the work of making it useful in your facility can cost as much as, or more than, the robot itself. Persistence Market Research's estimate that a mid scale warehouse AMR deployment may require 500,000 dollars to 2 million fits right in. It's like buying a car and discovering you also have to pay for the garage, the license, the driver training, the insurance, and the road. The sticker price is the beginning of the budget. This hits smaller companies hardest, and Fortune notes that complexity adds engineering effort, testing, and failure points at every interface with warehouse management systems, industrial controls, and cybersecurity. One response is robot as a service, where customers rent capacity on recurring terms instead of owning assets. Fortune cites IFR reporting growing adoption of that model. It can lower the entry cost, though it moves the capital burden back onto the robot company's balance sheet.

Zach Martin

Here's where I want to slow down on a number everyone loves quoting, the size of the physical AI market, because the sources disagree dramatically. Fortune Business Insights values the global physical AI market at 42.06 billion dollars in 2026, growing to 294.07 billion by 2034, a compound annual growth rate of 27.5 percent. It puts the 2025 figure at 33.42 billion. Persistence Market Research values it at 87.5 billion dollars in 2026, reaching 741.6 billion by 2033, at 35.7 percent. That means for the same year, one firm says roughly 42 billion and the other says roughly 87.5 billion. The second is more than double the first. Both agree hardware is about 55 percent of the market, and both are forecasts, not measurements. A likely reason for the gap is definitions. Fortune explicitly includes robotics, autonomous vehicles and machines, and vision AI and smart spaces. I can't tell you precisely how Persistence draws its line. So when somebody gives you one definitive total addressable market for physical AI, treat it with suspicion. The honest answer is that it's somewhere in a wide range, and the range depends on what you decide to count.

Zach Martin

And then there are valuations. Figure is valued at 39 billion dollars and NEURA at about 7 billion, on limited deployment fleets. Unitree is the only transparent baseline we have. A public company, audited revenue of roughly 1.7 billion yuan, about 240 million dollars, and a market value the brief describes as in the tens of billions. Dealroom offers a sentence about AI companies in general that I think applies. It says the valuations arrive well ahead of the businesses, and that at those figures a 10 billion dollar valuation is a forecast rather than a multiple of anything. It also observes that many of these companies were priced on their founders' records rather than on anything they had shipped. That is not an accusation of wrongdoing. Forecasts can come true. Plenty of valuations that looked absurd turned out to be cheap. But a forecast carries a burden of proof. For these companies the proof is straightforward to describe and hard to deliver. Customers paying, robots staying employed, and each machine earning back its cost. If a robot sits idle half the day, the math changes.

Zach Martin

Let me give you the signposts I'd watch over the next several quarters, so you've got a scoreboard rather than just a mood. One. Do pilots become recurring deployments in the thousands, measured by units in service, not units announced? Two. Utilization. Fortune says utilization, reliability, interoperability, and recurring operating value will matter more than prototype novelty, so ask how many hours per day a robot actually works. Three. Does the middle of the funding market recover? If rounds between 20 and 50 million dollars stay at 3 percent of capital, the pipeline of future leaders is thin. Four. How do the public tests go? Unitree's trading, UBTECH's ability to keep raising, and what happens when Agility goes through its SPAC. Five. The strategic money. Do the carmakers and component suppliers who invested also place repeat orders? And six, the source disagreements themselves. Keep a note of which numbers are closed equity and which are up to figures like NEURA's, and which market size forecast somebody is using. Careful bookkeeping is a competitive advantage.

Zach Martin

So let's pull it together. In 2026, humanoid venture funding hit 11.0 billion dollars through October 6, more than double the prior full year record. China holds 63 percent of it. A small number of mega rounds hold most of the dollars, and the middle of the market has thinned. Money is flowing to automotive plants, warehouses, defense, construction, datacenters, hospitals, and farms. Big Tech, NVIDIA especially, sits at the center of the stack. Public markets and a SPAC are about to grade the work. And the commercial evidence, while real, is small compared with the capital. My coaching takeaway is the one I'd give anyone in any field. Big moments are exciting, but compounding is what wins. The companies and the investors who last will be the ones who track the boring numbers, hours worked, cost per task, repeat orders, and keep improving them every month. A record funding year is a starting gun, not a finish line. I'm Zach Martin. This was Robot Capital on KMKT IR Hub Radio. Keep your discipline, keep your scoreboard honest, and I'll see you next time.