Humanoids Go Public: Inside Agility Robotics' $2.5B Listing

Zach Martin

Welcome to Robot Capital on KMKT IR Hub Radio. I'm Zach Martin, and it's Tuesday, October 6, 2026. I want to open with a promise I made last time. In the previous episode we walked through the first half of 2026 in physical AI venture funding: $47.4 billion across 521 deals, driven by giant rounds for Waymo, Anduril, Shield AI, and Saronic. And I left you with a caution. For all that private money, robotics still hadn't faced a real public market liquidity test. Well, the calendar caught up with the story. This morning Agility Robotics is hosting its Analyst and Investor Day, ahead of a planned listing through a merger with a special purpose acquisition company called Churchill Capital Corp XI. The deal values Agility at $2.5 billion. So that's where we start today, because it's the closest thing this industry has to game day.

Zach Martin

Here's how I think about the difference. A venture round is practice. Everyone is in the gym, the coach is optimistic, and the only people grading you are the ones who already bet on you. A public listing is game day. Strangers show up, they've paid for a ticket, and they don't care how hard you trained. They care about the scoreboard. So over the next hour I want to build the scoreboard together. We'll start with the Agility deal and what's actually inside it. Then we'll look at September's funding numbers, which tell a different story than the headlines about mega rounds. Then we'll follow the money across borders, because the geography of humanoid investment has flipped in a way that surprised me. After that we'll look at the big strategic buyers who've been quietly collecting pieces of the stack. And we'll finish with the least glamorous topic in robotics and probably the most important: gross margin. Why some robot companies can compound, and why others may just sell very impressive machines.

Zach Martin

Let's start with the structure of the Agility deal, because the mechanics shape everything else. Agility is merging with Churchill Capital Corp XI, which trades on Nasdaq under the ticker CCXI. After closing, the company is expected to list under the ticker AGLT. The pre money equity valuation is $2.5 billion. The expected gross proceeds exceed $620 million, and that number has two parts. About $420 million would come from the Churchill XI trust account, and that figure assumes no redemptions. The other $200 million is a PIPE, a private investment in public equity, committed at $10 a share. It's led by Foxconn, with institutional backers that include DCVC, NVIDIA, Amazon, and SoftBank Vision Fund 2. If you've never followed a SPAC, picture it like this. Someone builds a covered stadium and fills it with cash before any team has been chosen. The money comes first, and the company arrives later and moves in. That's useful for a company that needs capital, and it's also why the details matter so much.

Zach Martin

Now the part that has traders' attention. Shares of CCXI traded as high as $19.10 ahead of today's investor day. That's an 86.9 percent surge over the trailing 30 days. Keep that next to the PIPE price of $10 a share. The big institutions committed at $10. Anyone buying on the open market at $19.10 is paying nearly double that. I want to be careful about what that means and what it doesn't. A SPAC share price before the merger closes is a bet on the deal. It's trading the story, the scarcity, and the excitement of a pure play humanoid name arriving in public markets. It isn't a verdict on the business. We're still in the stretch where the stock is a story and the company is a spreadsheet, and the two haven't met yet. If you've ever watched a team's odds swing wildly the week before a championship, you know the price tells you about the crowd more than the players. Today is when the players get to speak.

Zach Martin

So what does the business look like? Agility says it has more than $300 million in multiyear contracted orders for its Digit v5 robot, across a pipeline of more than 30 enterprise customers. And I want to underline the qualifier that came with that claim: the orders are subject to milestone satisfaction. A contracted order that depends on milestones is a promise with homework attached. It's real, it's better than a press release about interest, but it's not cash yet. On the operating side, Digit has logged more than 65,000 operating hours across nine customer facilities, including commercial pilots with Schaeffler, GXO Logistics, and Toyota Motor Manufacturing Canada. Let's put that in human terms. 65,000 hours is about seven and a half years if one machine ran nonstop, every hour of every day. Spread across nine sites, that's roughly 7,000 hours per facility. That's meaningful evidence of robots actually working in real buildings. It's also a long way from a fleet of thousands.

Zach Martin

Here's another piece of arithmetic worth doing. $2.5 billion divided by $300 million in contracted orders is a bit over eight. Eight times contracted orders. That's not a revenue multiple, so please don't hear it that way. Orders aren't revenue, and these orders are conditional. But it gives you a sense of what the market is being asked to underwrite: not the past, but how fast that contract book converts into delivered robots, and then into repeat orders. I'll also note that Dealroom's data shows Agility having raised about $570 million in venture funding since 2020 at that $2.5 billion latest valuation. So the public listing isn't a rescue story in the sense of a company running on fumes. It's a company that wants a bigger war chest and a currency, public stock, for the next phase.

Zach Martin

The next phase is manufacturing. Agility's production is backed by its Salem, Oregon facility, called RoboFab, built for 10,000 units a year, with roughly 75 percent of parts sourced domestically in the United States. Let me put that capacity number next to another figure from the research. Global humanoid robot shipments reached only about 13,300 units in 2025, according to the research firm TSG Invest, which forecasts 250,000 units by 2030. So one factory in Oregon is designed to build a number approaching the entire world's humanoid shipments from last year. Capacity isn't output, and I'd never confuse the two. A factory sized for 10,000 units a year is a statement of intent and a bill of fixed costs. It's like signing a lease on a gym built for 10,000 members when you have a few hundred. It can be the right bet if the members are coming. If they're not, you're paying for empty floor space, and in a hardware business that cost doesn't go away.

Zach Martin

Now, why a SPAC at all? Why not a traditional IPO? The Robot Report and the law firm Mintz have both walked through this reasoning, and the argument has two halves. First, the traditional IPO window in 2026 has been congested by mega debuts. The Crunchbase data tells us SpaceX raised $75 billion in its June IPO at a $1.77 trillion valuation, the largest IPO of all time. When that's the show in town, a smaller company with a different profile has trouble getting attention. Second, pre revenue and pre profit robotics companies have difficulty clearing the hurdles that traditional underwriters set. Banks like revenue histories and predictable numbers, and a humanoid robot company in 2026 is selling the future more than the quarterly run rate. Andra Keay of Robots and Startups, who writes the monthly funding roundups I'll lean on in a minute, put the case for SPACs plainly. She noted they went through a bubble, but they remain a useful mechanism for companies that need to raise capital, don't want to dilute excessively, and don't yet meet Wall Street's IPO revenue markers.

Zach Martin

Now the other side of that coin, because a measured view requires both. The same analysis flags the known hazards of the de SPAC structure. First, redemption risk. Public shareholders in a SPAC can take their cash back out of the trust before the merger closes. The $420 million trust figure assumes none of them do. If many of them redeem, the trust arrives depleted. Second, reliance on the PIPE. When redemptions hurt, the PIPE becomes a backstop rather than pure growth capital, and in the worst case on paper, the floor under the deal is that $200 million, not the headline $620 million. Third, dilution. SPAC structures hand out shares to sponsors and other participants along the way, and existing holders can end up owning a smaller slice than the valuation headline suggests. And fourth, regulatory scrutiny. The SEC has tightened SPAC disclosure guidelines, which raises the bar on how companies describe their projections. So there's a tension in the story. Keay says SPACs help avoid excessive dilution, while the analysts flag dilution as a hazard. Both can be true, depending on how a particular deal is built, which is exactly why today's disclosures matter.

Zach Martin

Two more pieces of recent news fill in the picture. On September 29, Agility announced its incoming public board members. And on October 1, it announced a strategic safety partnership with FORT Robotics. According to Keay's roundup, FORT Robotics is also headed toward public markets via a reverse merger, so two companies in the same corner of the industry are going public at the same time through the same type of structure. The safety partnership is worth a moment of thought. A humanoid robot that works near people is a different kind of risk than a robot arm behind a cage. Anyone who's going to put a machine in a building with employees will ask about safety before they ask about price. My interpretation, and I'll flag that it's mine, is that a safety partnership announced days before an investor day is a signal about which questions management expects to get. When you hire a safety specialist right before the biggest scrutiny of your life, you're telling the room you know what they're going to ask.

Zach Martin

I'm recording this while the investor day is still underway, so I can't tell you what management is saying right now. What I can do is tell you what I'd listen for. First, conversion. How do those $300 million in contracted orders turn into delivered robots, and on what schedule? What are the milestones, and who decides they've been met? Second, depth per site. Nine customer facilities is a start, but does a site that's running a pilot expand to ten robots, then fifty? The business gets interesting when a customer reorders without being asked. Third, use of proceeds. Is the money going toward the Salem factory, toward software, toward safety certification, toward the sales team? And fourth, the unit economics. What does it cost to build a Digit, what does a customer pay, and what's left over? We'll come back to that last question at the end of the show, because it's the one that separates a good demo from a good business. For now, remember the coaching principle: game day doesn't build fitness, it reveals it.

Zach Martin

Let's zoom out from one company to the whole monthly pulse. Andra Keay's Robots and Startups newsletter published the September 2026 funding tally on October 3. The headline: 167 funding rounds for a total of $4,862,394,030. Call it $4.86 billion. That's less than August or July in dollars, but it's more rounds. And Keay adds two caveats that I want you to hold onto. About a third of the entries report no funding amount, and large late stage rounds always skew the totals. So this isn't a perfect measurement. It's a thermometer, not an MRI. But even a thermometer tells you something. When the dollar total falls and the deal count rises, it usually means the biggest checks were less dominant that month, and more companies got funded. It's like a month where the star players rest and the bench scores more points.

Zach Martin

Let me do some rough math, flagged as rough. $4.86 billion over 167 rounds averages about $29 million per round. But if roughly a third of the rounds have no reported amount, the reported ones are probably closer to $43 million on average. Now, the named deals in Keay's list add up to about $1.09 billion, a little over a fifth of the month's total. One deal, Mach Industries at $600 million, is about 12 percent of the entire month's dollars all by itself. So even in a month that looks dispersed, a single defense round carries real weight. I'll also say Keay lists these as notable deals, not a complete list, so the rest of that $4.86 billion is spread across rounds I can't name for you. The picture is a long tail with one fat shoulder, and the shoulder is defense.

Zach Martin

Let's walk through the clusters, because the variety is the story. Defense and aerial infrastructure first. Mach Industries raised $600 million for drones and defense infrastructure. Poseidon Aerospace raised $60 million for cargo drones. Sunflower Labs raised $15 million for drone security systems. Then industrial and field automation. Maven Robotics raised $100 million for industrial robots. Planted raised $31 million for solar installation robots. Prime Minute raised $15 million for disaster response robots. Then something I find fascinating: datacenter robotics. Watney Robotics raised $80 million for datacenter robots. Medical: Zap Surgical raised $11 million for radiosurgery robotics aimed at brain tumors. And the picks and shovels layer: Lyte AI raised $165 million for integrated perception, and Robocurve raised $10 million for open benchmarks for robotics. That's ten companies across defense, solar construction, disaster response, data centers, medicine, and measurement. None of them is a humanoid in a warehouse, which is the image most people have when they hear the word robotics.

Zach Martin

Why does the datacenter deal catch my eye? Because of who's paying attention to data centers right now. In the Crunchbase piece from earlier this year, investor Joe Fath of Eclipse Capital described funding moving toward energy, robotics and autonomy, inference, chips and compute, and data center infrastructure, all as parts of one physical AI picture. His firm describes physical AI as intelligence embedded in systems that perceive, reason, and act in the real world, and it invests on what he calls the shoulders rather than the head. Chips, compute, energy, data centers on one shoulder, and companies applying AI to build new businesses in the physical world on the other. A company building robots to work inside data centers sits right where those two shoulders meet: it serves the AI buildout while being a physical AI business itself. I can't tell you from the sources whether Watney's approach works, because the roundup gives me only a short description. But I can tell you why a funder would find the target attractive.

Zach Martin

There's also the matter of a rumor at the other end of the size spectrum. Keay notes that Field AI is apparently raising a $700 million venture round at a $10 billion valuation, according to Business Insider. I'll stress the word apparently. This is reported, not announced. But if it closes, it would add another giant valuation to a field that already has Skild at $14 billion and Figure at $39 billion, and we'll get to both. What I want you to notice is the barbell. At one end, a few companies are raising hundreds of millions at valuations in the billions. At the other end, there's a thick middle and long tail of rounds in the tens of millions. September's data is the tail getting thicker at the same time the head took a breath.

Zach Martin

Now a caution about reading too much into the word dispersion. It's tempting to say capital is flowing into seed stage companies everywhere. I don't have a stage breakdown for September, so I can't confirm that. And one background report, Bricks and Bytes' Physical AI and Robotics 2026, describes a bifurcation in capital markets between mega rounds and a flat early stage market. That report says global robotics investment hit a record $20.4 billion in 2025, with about $233 billion in private unicorn value across physical AI. So there are two pictures. One says money is spreading into many applications. The other says the very earliest stage hasn't been getting the same lift as the giant rounds. They can both be partly right. September's 167 deals show breadth in the middle of the market, but I wouldn't tell a founder with a seed deck that the money is raining down. Treat the data like a weather report: useful, local, and not a promise.

Zach Martin

Let me put September into the larger context from this week. Crunchbase published its Q3 numbers on October 5. Global venture funding across all sectors totaled $159 billion in the third quarter, with close to 6,000 startups funded, and the third quarter posted a record count of billion dollar rounds as the global AI race heats up. Different scorekeepers use different rules, and Crunchbase's physical AI category includes autonomous vehicles, aerospace, drones, industrial automation, and sensors, while Keay's monthly tally is specifically robotics. So please don't divide one by the other and call it a market share. But the direction of the water is the same. AI capital is abundant. The question for robotics is whether it flows to companies that can show production milestones, because that's exactly what Fath said is happening. Funding is shifting away from experimentation and toward companies that can hit production milestones, land customers, and scale efficiently.

Zach Martin

Ryan Ziegler, a general partner at Edison Partners, offered a useful frame for why these varied deals make sense together. He told Crunchbase News that physical AI is the convergence of software, hardware, sensors and IoT, and services across many real world applications. And he made a point I keep returning to: many of these companies look like vertical software businesses, with attractive unit economics, large deal values, and multiyear deployments. The hardware, in his words, can be the distribution model for creating a data intelligence flywheel. In plain English: you sell or deploy a machine, and the machine collects data, and the data makes the software better, and the better software makes the next machine more valuable. He also compared the cost decline to what cloud infrastructure did for software as a service, noting that compute, foundation models, simulation, training data, and sensors have all become cheaper or more accessible. Even phones now carry LIDAR scanners, he pointed out. If building gets cheaper, you'd expect more companies to form, and more companies to get funded. That's consistent with a fatter tail in September.

Zach Martin

Here's the coaching takeaway for this section. When you watch your own business, or when you watch a portfolio, a single big number is a trap. The better habit is to track the shape. How many deals, how big the largest one is, how many distinct industries are represented, and how much of the total is unknown. September's shape says breadth is growing even as the total dips. It says defense carries the biggest single weight. It says an old problem, measuring who's actually doing what in robotics, is getting its own funding, because Robocurve's $10 million is for open benchmarks. That's a small check, but a telling one. When an industry starts paying for ways to measure itself honestly, it's moving from demos toward discipline. Small improvements, compounding. That's the pattern I want to see more of.

Zach Martin

Now let's follow the money across borders, because this is where the data surprised me most. Dealroom, the European data platform, published a humanoid robotics set of charts. Venture investment into humanoid robotics startups has reached $8.7 billion in 2026, and I want to flag the qualifier: that's year to date through July 22. It's already nearly double 2025's full year record of $4.7 billion. Think about the trajectory. In 2020, the number was $61 million. In 2021, $645 million. 2022 dipped to $453 million, 2023 was $547 million, and 2024 reached $1.5 billion. Then $4.7 billion in 2025, and $8.7 billion in less than seven months of 2026. That's the kind of curve you'd normally see in the middle of a bubble, or in the early part of a real platform shift. I can't tell you which, and I'd be skeptical of anyone who says they can. But it's undeniable that humanoids have become a destination for capital.

Zach Martin

Now the geography. By company headquarters, China accounts for about 65 percent of humanoid venture funding in 2026, up from 25 percent in 2020 and 46 percent in 2025. The United States share has fallen to about 15 percent, down from 54 percent in 2024 and 45 percent in 2025. Europe sits at 18 percent, and the rest of the world around 2 percent. Let me translate those shares into dollars, using Dealroom's chart. China went from about $2.2 billion in 2025 to $5.6 billion so far in 2026. Europe went from $274 million to $1.6 billion. The United States went from $2.1 billion to $1.3 billion. So this isn't only a shift in share. In dollar terms, the U.S. total is actually lower so far this year than the entire prior year, while China's total has more than doubled. That's a real reallocation. It's like the center of gravity in a sport shifting to a different league, and the scouts following it.

Zach Martin

But before we conclude that America is losing, let's look at the timing, because it matters. Dealroom's 2026 numbers cover announced rounds this year. Figure AI's big Series C, more than $1 billion at a $39 billion post money valuation, was announced in September 2025. So the largest U.S. humanoid round of this cycle landed in last year's column, not this year's. That alone depresses the 2026 U.S. share. It doesn't explain everything, since China and Europe are clearly accelerating, but it's a reminder that annual snapshots can be lumpy. The same holds in reverse. Dealroom's figures are for announced VC rounds into venture backed companies carrying its humanoid sector tag. Skild AI's $1.4 billion Series C was announced in January 2026, and it doesn't appear on Dealroom's list of the largest humanoid rounds. My interpretation is that a company building a brain for many robot bodies sits outside that tag, but I can't verify that from the chart. The lesson is to know exactly what a category includes before quoting a number.

Zach Martin

Let's name the biggest rounds of 2026 on Dealroom's list. The largest is a $1.4 billion Series C for Germany's Neura Robotics. Neura has raised about $1.7 billion in total, at a $7 billion valuation. Next, AI2 Robotics raised $735 million in a late venture round. Apptronik, the Austin company, raised a $520 million Series A. Rhoda AI raised $450 million in a Series A. Galbot raised $344 million in a late venture round. Walden Robotics raised a $300 million seed round, and I'll pause on that one because a $300 million seed is a striking phrase. And Galaxea AI raised $280 million in a Series B. Notice the mix. Germany, China, and the United States all appear. The money is global, even though the Chinese share is the biggest by far. And Dealroom notes that a handful of companies hold most of the capital: Figure and Neura have each raised over $1.5 billion since 2020, and twelve companies now sit on war chests of $380 million or more.

Zach Martin

Here's an exercise I find clarifying. Compare how much each company has raised with the valuation that Dealroom lists. Figure: $1.9 billion raised, a $39 billion valuation. That's about twenty times the capital raised. Neura: $1.7 billion raised, $7 billion valuation, about four times. Apptronik: $950 million raised, $5.5 billion valuation, about six times. Galbot: $964 million raised, $2.9 billion valuation, roughly three times. AI2 Robotics: $918 million raised, $2.9 billion valuation, again about three times. Now, I'm not telling you which of these is right. A high ratio might mean investors see an enormous future, or it might mean the price is running ahead of the proof. A low ratio might mean a bargain, or it might mean the market doesn't see the same potential. What the pattern does show is that Figure is priced very differently from nearly everyone else, and the Chinese companies on the list are priced at roughly three times the money they've taken in. When one asset sits that far off the pack, you should ask exactly what the market is paying for.

Zach Martin

What is it paying for in Figure's case? Let me share what the sources say, and then separate fact from reading. The facts: Figure was founded in 2022 by Brett Adcock. It raised $675 million in February 2024 at a $2.6 billion valuation from a group including Jeff Bezos, Microsoft, Nvidia, Intel, and the startup funding arms of Amazon and OpenAI. In September 2025, it raised more than $1 billion at a $39 billion valuation. By April 2026, according to reporting summarized on Wikipedia, the company had delivered over 350 Figure 03 robots and increased production to one robot every hour. In May, a livestream showed Figure 03 units sorting packages, and as of the early morning of May 20, the stream indicated more than 204,000 packages sorted in more than 163 hours, according to Forge Global. And in a 10 hour comparison, the robots performed 98.5 percent as well as a person. Those are striking demonstrations. They're also exactly the kind of demonstrations a company chooses to show you. A livestream proves a robot can do a task repeatedly. It doesn't tell you the unit cost, the uptime across a year, or what happens when the package is the wrong shape.

Zach Martin

Wikipedia also records a piece of the Figure story that investors should keep in view. In November 2025, the company's former head of product safety sued, saying he was fired for raising a concern that the robots were strong enough to fracture a human skull. That's an allegation in a lawsuit, not a finding, and I don't have the company's response in the material I've got. I mention it because it connects to something we discussed with Agility: the safety partnership with FORT Robotics. In a humanoid business, safety isn't a side topic. It's a cost line, a regulatory topic, a customer procurement hurdle, and a reputational risk, all at once. Whichever company you favor, ask how it talks about safety when nobody's forcing it to.

Zach Martin

Let's talk about what's driving the Chinese share, with the caution that the sources I have describe the amounts, not the reasons. Dealroom lists AI2 Robotics at $735 million, Galbot at $344 million, Galaxea AI at $280 million in a Series B, plus companies with accumulated funding like Robot Era at $811 million, Giga AI at $591 million, LimX Dynamics at $496 million, and PsiBot at $380 million. That's a deep bench. The Dealroom chart shows China at about 25 percent of humanoid venture funding in 2020 and 65 percent in 2026. I won't speculate about the policy or industrial reasons, because the evidence I have doesn't cover them, and I'd rather tell you what I don't know than invent a story. What I can say is that if you're an investor or an operator who thinks about supply chains, hardware talent, and manufacturing scale, you can't treat the humanoid race as a mostly American contest anymore. The numbers don't support it.

Zach Martin

One more datapoint to keep our feet on the ground. Remember that TSG Invest, the research firm, cites about 13,300 humanoid shipments globally in 2025, forecast to reach 250,000 by 2030. And look at the forecasts for the market size. Goldman Sachs projects the humanoid robot market could reach $38 billion by 2035. Morgan Stanley projects $5 trillion by 2050. Those are different dates, so they aren't strictly a contradiction, but the gap between them tells you how uncertain the whole exercise is. When forecasters from respected institutions are that far apart, the honest takeaway isn't that one is right. It's that nobody really knows the shape of this market yet. And meanwhile, $8.7 billion has already gone in this year alone through July. That's a lot of capital chasing a market whose size is a range of orders of magnitude. My view is to respect the energy, and stay disciplined about the exits. Money in is not money out.

Zach Martin

Now let's talk about the buyers, because while venture funds write checks into startups, a different group has been quietly assembling a stack by acquisition. September had a cluster of these deals, and when you line them up, they tell a coherent story. First, Qualcomm acquired PickNik Robotics, according to Keay's roundup, to expand the open source robotics navigation and manipulation framework. Second, SoftBank Group is acquiring full ownership of the Robotics and AI Institute, known as RAI, from Hyundai, a transaction that's pending a U.S. government review of foreign investment. Third, Cognex acquired the RealSense 3D depth sensing business, for approximately $500 million in cash. RealSense was Intel's business. Three different buyers, three different layers. Software, research, and sensing. If you were building a team, that's like drafting a coach, a scouting department, and a pair of eyes.

Zach Martin

Let me take them one at a time and keep the facts separate from my reading. PickNik is known for work around open source frameworks for robot motion and manipulation. The research I have says Qualcomm is acquiring it to expand that framework. The logical reading, which is mine, is that a chip company wants the software tooling that makes robots run well on its chips, because a processor that no developer can easily program is a hard sell. I want to flag that this is interpretation. The facts are the acquirer, the target, and the stated purpose. Now Cognex and RealSense. Cognex is a machine vision company, and about $500 million in cash for a 3D depth sensing unit consolidates vision hardware under a firm whose whole business is making machines see. Again, the numbers I can verify are the price and the parties. The reasoning about why the depth camera matters for robots is intuitive: you can't grasp what you can't measure.

Zach Martin

The SoftBank piece is the most interesting to me because it fits a pattern. SoftBank is acquiring full ownership of RAI from Hyundai, pending that review. And it follows a much larger move: SoftBank's $5.375 billion acquisition of ABB's robotics division. And SoftBank has also led three consecutive rounds in Skild AI, the robot brain company, according to the TSG Invest analysis. Put those together. SoftBank owns or is acquiring industrial robot bodies through ABB's robotics division, a research institute through RAI, and holds a major position in a company building a general purpose brain. That's a thesis assembled in public: that software intelligence can unify diverse hardware. TSG Invest made that argument directly. I'd describe it as a reasoned interpretation by an investment research firm, not a statement from SoftBank. But the pieces are real, and the dollars are large. It's a bit like buying the stadium, the training facility, and a share of the team's top coach.

Zach Martin

Let me add the Mobileye deal from earlier in the year, because it belongs in this pattern. According to Crunchbase, Mobileye acquired Mentee Robotics, a Tel Aviv humanoid startup, for roughly $900 million, a transaction the company explicitly tied to its push into physical AI. That's a company with deep roots in vehicle perception deciding that its next growth path runs through humanoid robots. And notice the broader exit picture from the same Crunchbase report. Exits in 2026 have been concentrated in aerospace, defense, and drones more than in robotics proper. SpaceX raised $75 billion in its June IPO at a $1.77 trillion valuation. HawkEye 360 raised $416 million in its public debut, and the drone maker Aevex raised $320 million. The Bricks and Bytes report says something related about M&A: defense robotics took four of the six largest M&A exits above $250 million. So the pattern of liquidity so far is that defense and aerospace have had the exits, while humanoid robotics has had the valuations. Today's Agility event is part of the attempt to change that.

Zach Martin

What should a thoughtful observer take from the shopping spree? First, the buyers are mostly large companies with existing businesses to protect or extend. Qualcomm sells chips. Cognex sells vision. SoftBank is a giant investor that's now an owner. Mobileye sells automotive perception. None of them is a pure play startup. They're adding robotics because it complements something they already do, which is a pretty different motive than a venture fund looking for a ten times return. Second, the prices vary widely. $500 million for a sensing business, $900 million for a humanoid startup, $5.375 billion for a robotics division at an established industrial company. That range reflects what each asset is: a component, a startup with promise, and a business with customers and revenue. Third, notice that nobody here bought a company for an eye popping multiple of nothing. The sensing business and the robotics division are real operations. When strategic buyers pay, they pay for things they can integrate.

Zach Martin

Another thread worth pulling is how often the same names show up on the investor lists. For Agility's PIPE, the institutional backers include NVIDIA, Amazon, and SoftBank Vision Fund 2. For Skild AI, the investors include SoftBank, NVIDIA, Bezos Expeditions, Sequoia, Lightspeed, Samsung, LG, Schneider Electric, and Salesforce Ventures. For Figure's Series C, the investors included Brookfield Asset Management, Intel, Macquarie Capital, Nvidia, Parkway Venture Capital, Qualcomm, Salesforce, and T-Mobile. Nvidia appears in all three. Now, TSG Invest makes a sharp observation about this: Nvidia's own Isaac GR00T foundation model competes with Skild, yet Nvidia participated in Skild's funding, effectively hedging its position by backing what may become the dominant third party robotics intelligence layer. That's the analysis of an outside firm, not Nvidia's own statement. But the principle is easy to understand. If you sell the picks and shovels to every miner, you want a stake in each claim. It's a strategy of owning the ecosystem rather than picking one winner.

Zach Martin

There's a flip side, and measured thinking demands we face it. When the same strategic investors back competing companies, and when those same companies buy chips and compute from those investors, you get circular flows of money. I'll give a concrete example from the Figure story, with the caveat that I'm relying on a Forbes headline summarized in the reference material. In September, Figure AI announced it would spend upwards of $3.5 billion on Nvidia GPUs by next year. Figure has raised about $1.9 billion. So the compute commitment is nearly double all the capital the company has raised. A commitment isn't the same as cash paid up front, and I don't know how it's structured or financed, so I'm not suggesting anything is wrong. But it does show how much of the money flowing into physical AI is expected to flow right back out to the compute suppliers, some of whom are also investors. When you assess a round, follow the money one step further than the headline. Ask who ends up with it.

Zach Martin

Let me pull one more strategic thread. Skild's investor mix includes Samsung and LG for consumer robotics adjacency, Schneider Electric for industrial automation distribution, CommonSpirit for healthcare, and Salesforce Ventures for enterprise software integration. TSG Invest reads these as potential distribution channels into distinct end markets. I like that framing, with a caution that it's an analyst's reading of what the investors might bring, not a list of signed commercial agreements. In my experience watching businesses grow, distribution beats brilliance more often than people like to admit. A good product with no path to customers stalls. A decent product with a strong channel compounds. So when you see strategic investors who can open doors, that matters. When you see strategic investors who just want exposure, it matters less. You can't always tell the difference from the outside, which is why the follow up questions are so valuable.

Zach Martin

We've covered who's buying and who's funding. Now the hardest question in this whole space: how do these companies make money that lasts? Let's start with Skild AI, because it represents the cleanest version of the software thesis. Skild was founded in May 2023 by Deepak Pathak and Abhinav Gupta, two Carnegie Mellon professors who previously worked at Meta's FAIR research group. It's headquartered in Pittsburgh. It builds the Skild Brain, a general purpose foundation model meant to control many kinds of robots, from humanoids to quadrupeds to robotic arms, delivered through a cloud platform and API. Skild doesn't sell robot hardware. In January 2026, it closed a Series C of approximately $1.4 billion at a valuation above $14 billion, reported to be the largest robotics AI funding round on record, bringing total funding to over $2 billion. And the company reportedly grew from zero to about $30 million in revenue in a few months during 2025. A quick note on sourcing: this comes from TSG Invest, a firm that sells access to private market opportunities and says its page isn't investment advice, so I'd treat the revenue figure as a company reported number, not an audited one.

Zach Martin

Do the math the way TSG did. $14 billion divided by $30 million is about 467 times revenue. That's an enormous multiple, and TSG itself calls it among the highest of any venture backed company in any sector. What would justify it? The bull case is that the brain is horizontal, so every deployment on any robot body generates data that improves the model for all customers, a compounding flywheel. NVIDIA testing, according to TSG, showed the Skild Brain can run on hardware costing $4,000 to $15,000, versus $250,000 or more for traditional custom systems. If true at scale, that changes who can afford a robot. The bear case is just as clear. Simulated physics can't fully capture the real world. Production reliability across thousands of different environments is unproven. Competition includes Physical Intelligence, Nvidia's own Isaac GR00T, and vertically integrated humanoid makers like Figure, Tesla Optimus, Apptronik, Agility, and Boston Dynamics. And if the manufacturers choose to integrate vertically, as Tesla is doing with Optimus, the market for a horizontal brain could shrink. I don't know which case wins. But I know what I'd monitor: revenue growth that doesn't depend on a few large customers, and evidence that the flywheel improves performance in the field.

Zach Martin

Now let's talk about the margin problem, because it's the thread that connects Skild's software story to Agility's hardware story. The Bricks and Bytes report puts a number on it: a 37 to 41 percent gross margin ceiling that separates hardware vendors from compounding physical AI businesses. If gross margin is a new term for you, it's simply what's left from each sale after the direct cost of making and delivering the product. A company that sells a machine for 100 dollars and spends 60 dollars to build and ship it has a 40 percent gross margin. That 40 cents on the dollar has to pay for research, sales, support, and everything else, and it has to leave something over. Contrast that with the benchmark in the same report: Intuitive Surgical, the surgical robot company, has a 67 percent gross margin. How? Because it earns not only from selling the system but from recurring revenue around it. The lesson, per the report, is that physical AI companies need to capture software licensing, recurring robots as a service revenue, or consumables and services layers to justify software like venture multiples. Pure hardware sellers cannot compound at venture scale.

Zach Martin

Let me make that concrete with an analogy I like. Think of a gym. A gym that sells you a treadmill once makes a one time profit and needs a new customer next month. A gym with a membership collects every month, and the equipment is just how it earns the right to keep you. Robots as a service works like the membership model. Instead of a customer buying a robot outright, which is a big capital expense, the customer pays a recurring fee for work performed. The Bricks and Bytes report frames it as a comparison between robots as a service and a capital expenditure model, and argues the business model matters more than the hardware. The tradeoff is real, though. Under a service model, the robot maker carries the hardware on its balance sheet and waits to be paid back over time. That's heavy working capital. So a recurring model improves the quality of revenue, but it can strain cash. This is why I keep pointing back to unit economics. The right question isn't how many robots shipped. It's what each robot earns, over what period, at what margin, with what maintenance.

Zach Martin

Figure's recent moves show a company experimenting along exactly this axis. In May, Adcock said the company would soon lease its robot for home use for around $600 a month, according to reporting summarized in the reference material. That's a membership model in a home setting. In late August, Figure launched a gig economy platform called Index, designed to capture work data from humans wearing sensors, expected to cost $1 billion over the next year, per Forbes. Think about what that is. Figure is paying people to generate the training data that makes the robots smarter. And the Bricks and Bytes report points to a deep structural issue behind this: a roughly 10,000 times data gap between robotics datasets and modern large language models. Language models learned from the internet. Robots have no internet of physical experience to learn from. So whoever builds the biggest, best dataset of physical work may own a lasting advantage, and that's an expensive bet. Figure is spending a billion dollars to try to buy that advantage.

Zach Martin

Now let's connect that to Agility. When Agility's management speaks today, I'd love to hear how much of its revenue model is selling robots and how much is recurring. I'm not claiming to know the answer, because the material I have doesn't spell it out. But the framework tells me what to ask. If Agility's revenue is mostly hardware sales, then the 37 to 41 percent ceiling is the number to hold against it. If a large part is recurring service, software, or fleet management, the story improves. Remember too the $300 million in contracted orders, subject to milestones, and Digit's 65,000 operating hours with Schaeffler, GXO, and Toyota Motor Manufacturing Canada. The Bricks and Bytes report mentions Digit moving more than 100,000 totes at GXO, which suggests there's real work being done. Whether that work earns a margin that lets the company compound is the thing public investors will be paid to figure out.

Zach Martin

I want to offer a counterweight to the humanoid excitement, because some of the clearest revenue in physical AI right now doesn't have legs. The Bricks and Bytes report argues that autonomous heavy equipment is the investable story in construction while humanoids remain the camera ready one. It cites Bedrock at unicorn valuation, AIM with four years of commercial operations, and Crewline with $26 million of contracts already booked. It says construction tech venture capital alone broke records at $2.6 billion. And it describes a locomotion first wedge that does not wait for humanoid manipulation. In plain terms: machines that drive, dig, and haul can be useful without solving the hardest problem in robotics, which is dexterous manipulation of arbitrary objects. The report also lists four reasons it believes humanoids can't deploy on live construction sites today. That's an opinion from a single report, and humanoid makers would argue the timeline is shorter. But as an investor's habit, I like it: look for the version of the technology that earns today, then watch for the version that might earn tomorrow.

Zach Martin

So let's pull the whole episode together, the way a coach would in the locker room. This week, the public market test finally began. Agility's $2.5 billion de SPAC is the first big look at how Wall Street prices a pure play humanoid company, with a stock that ran up 86.9 percent in 30 days before the investor day and a PIPE struck at $10. September showed 167 rounds and $4.86 billion, a broader mix of defense, industrial, datacenter, medical, and perception companies, with the dollars down and the deal count up. Humanoid venture money reached $8.7 billion through July 22, and its center of gravity moved to China at about 65 percent, with Europe's Neura raising the biggest round of the year so far. Strategic buyers, Qualcomm, SoftBank, Cognex, Mobileye, have been buying layers of the stack. And underneath all of it sits a quieter question about whether the revenue is the kind that compounds.

Zach Martin

If I had to give you three habits for following this sector without losing your footing, they'd be these. One: follow contracts to cash. A pilot is a hello, a contracted order is a handshake, delivered robots are a signed lease, and recurring revenue is a relationship. Know which one you're looking at. Two: respect gross margin. Ask what each machine earns after the cost of building and serving it, and whether software or services sit on top. The 37 to 41 percent ceiling for pure hardware and Intuitive's 67 percent aren't laws of nature, they're a benchmark from one report, but they're a useful ruler. Three: know whose money it is and where it goes next. Strategic investors, chip suppliers, and compute vendors often sit on both sides of the table, so follow the flow one step past the headline. None of this is exciting in the way a robot doing a backflip is exciting. But progress in any discipline, in business or in training, tends to come from steady, boring habits compounded over time.

Zach Martin

A closing note on caution, because it belongs here. Everything in this episode is educational and drawn from public reporting, research firms, and the sources I named. It's not investment advice, and several figures, like Skild's revenue, Agility's contracted orders, and Field AI's rumored round, are reported or company stated numbers I haven't independently verified. A SPAC stock price can swing hard either way around a merger, and private valuations are marks, not prices at which you can sell. If you're deciding what to do with real money, talk to a qualified professional who knows your situation. The best position in a hot sector is usually the patient one: curious, informed, and unwilling to chase. I'll be watching how Agility's story develops through closing, how September's breadth shows up in October's numbers, and whether the strategic buyers keep collecting pieces. Thanks for spending the hour with me. This is Zach Martin, and this has been Robot Capital on KMKT IR Hub Radio. Keep your habits steady, keep your questions sharp, and I'll talk to you next time.