Physical AI’s Mega-Round Moment

Father Zachary

Welcome to ROBOT CAPITAL, the business, investment, and technology of the robotics revolution. I'm Zachary. It's early October 2026, and this episode is the capital intelligence briefing: the mega rounds, the public filings, the acquisitions, and the investor theses, brought into one view. Our rule hasn't changed since Episode 1. Follow the companies. Follow the technology. Follow the capital. And our discipline hasn't changed either. We don't grade a company on its demo video. We grade it on four things: customer purchase orders, deployed unit economics, hardware and software gross margins, and how long the balance sheet can last. Those are the same working capital realities we walked through in Episodes 2 and 3 on robots as a service. Today we apply them to the biggest capital flows of the year.

Father Zachary

Start with one number. In the first half of 2026, global venture funding for physical AI totaled $47.4 billion across 521 deals. That's Crunchbase data, published by Mary Ann Azevedo. Now give that number some context. In the second half of 2025, physical AI startups raised $12 billion across 470 deals. So the first half of this year is almost four times the prior half. Against the first half of 2025, which saw $26.4 billion across 436 deals, it's up nearly 80%. Notice what moved and what didn't. The deal count rose modestly, from 436 to 521 against last year's first half. The dollars rose dramatically. More money is chasing roughly the same number of companies, and that tells you the checks are getting bigger, not that the field is getting wider.

Father Zachary

Here's the comparison I'd put on a slide. Across the three full years from 2022 through 2024 combined, venture investors put $41.9 billion into physical AI. Crunchbase notes that's still several billion dollars less than what was raised in the first six months of 2026 alone. Three years of capital, surpassed in half a year. The Wall Street Journal saw the shift building early. In a March 10 piece, WSJ Pro reporter Jon Leckie wrote that through February 25, global venture funding for physical AI had reached $26.7 billion, including $16 billion raised by Alphabet's autonomous vehicle company Waymo, per Crunchbase. That total, the Journal noted, was already more than any annual investment before 2025. So by late February, one quarter into the year, the sector had out-raised every full year before 2025.

Father Zachary

A definitional note, because it changes how you read every statistic in this episode. Crunchbase defines physical AI to include robotics, autonomous vehicles, aerospace, drones, industrial automation, and sensors. That's a wide net. When you hear forty seven billion, you are not hearing forty seven billion for humanoids, or for warehouse arms, or for the robots we usually cover on this show. You're hearing a basket that includes a robotaxi operator, defense drone makers, and space intelligence companies. That doesn't make the number less real. It means the number measures the physical AI economy, and the robotics slice has to be pulled out by hand. Which is what we'll do for the rest of this briefing.

Father Zachary

Now the concentration, which is the real story inside the headline. Waymo's $16 billion Series D, raised in February, accounts for nearly one third of all venture dollars in the half, according to Crunchbase. It was led jointly by Alphabet, Dragoneer Investment Group, DST Global, and Sequoia Capital, at a $126 billion valuation. One company, one round, a third of the sector's first half. And the leaders didn't stop there. Let's add the other mega rounds that Crunchbase highlights. Anduril Industries raised $5 billion in May at a $61 billion valuation. Shield AI raised a $2 billion Series G in March. Saronic raised a $1.75 billion Series D, also in March. Add those four rounds together: sixteen, plus five, plus two, plus one point seven five. That's $24.75 billion.

Father Zachary

Here is our arithmetic, not Crunchbase's, so treat it as interpretation. Twenty four point seven five billion against forty seven point four billion is a little over half, roughly 52%. Four rounds, in two categories, autonomous driving and defense autonomy, account for about half of all physical AI venture dollars in the first half of 2026. If you strip out Waymo alone, the remaining roughly $31.4 billion is spread across 520 deals, an average of about $60 million a deal. Include Waymo and the average is about $91 million. For comparison, the second half of 2025 averaged roughly $25 million a deal on the same math. So even without the largest check of the year, the average round more than doubled. Averages hide the distribution, and we don't have the median here, so I won't claim the typical startup got a bigger check. What I can say is that the top of the market got dramatically richer, and the sector's total is heavily dependent on a handful of names.

Father Zachary

Look at the defense valuations, because they tell you what investors are underwriting. Anduril's $61 billion valuation is double the $30.5 billion it received less than a year earlier. Shield AI, based in San Diego, landed its $2 billion Series G co led by Advent International and JPMorgan Chase, and its valuation jumped to $12.7 billion. Saronic, the Austin company building autonomous sea vessels, raised $1.75 billion led by Kleiner Perkins, bringing total funding to around $2.6 billion and setting a $9.25 billion valuation, more than double its Series C level in 2025. Now apply our scorecard. In defense, the customer is a government, the purchase order is a contract, and the demand signal is an explicit budget line. That is a structurally different underwriting question from a humanoid that has to win over a plant manager one pilot at a time. I'd call that my interpretation of why defense capital moves faster. The sources report the valuations, not the reasons.

Father Zachary

Now liquidity, because a venture surge without exits is a paper surge. The exit side of 2026 has been real but narrow. SpaceX raised $75 billion in its June IPO at a $1.77 trillion valuation, which Crunchbase calls the clear outlier. Crunchbase's related coverage says SpaceX shares closed up 19% after what was described as the largest IPO of all time. Smaller public debuts followed: HawkEye 360, the Herndon, Virginia space intelligence company, raised $416 million, and Aevex, the Arlington, Virginia autonomous drone maker, raised $320 million. Crunchbase's own observation is that activity has been more concentrated in aerospace, defense, and drones than in areas like robotics. Hold on to that sentence. It's the hinge of this whole episode.

Father Zachary

On the acquisition side, the most notable deal Crunchbase points to is Mobileye's roughly $900 million purchase of Tel Aviv humanoid robotics startup Mentee Robotics, a transaction the company explicitly tied to its push into physical AI. Think about what that is. An established automotive technology supplier is buying its way into humanoids. For a robotics founder, that's the strategic acquirer pathway, and it's one of the few robotics exits in this dataset. So here is the picture. Private capital is flowing into robotics at record scale. The public exits so far are mostly space, defense, and drones. Robotics is waiting for its liquidity moment, and as we'll see in the next chapter, the first public tests are arriving in September and October, and they're not uniformly flattering.

Father Zachary

One more macro data point from The Robot Report. Robotics investments reached $4.9 billion in August 2026. That's a single month, the most recent full month in our data, and it is a useful reminder that outside the mega rounds, the flow is steady. Also, a note on how I'll handle the sources. When a number comes from Crunchbase, the Journal, or The Robot Report, I'll say so. When a figure is my own arithmetic, I'll tell you. And when a company makes a claim, I'll mark it as the company's claim. That's the anti demo discipline in practice. A valuation is an opinion with a price tag. A purchase order is a fact. So as we go through the rounds, keep asking: where is the purchase order, what is the margin, and how many quarters of cash are on the balance sheet?

Father Zachary

Let's go to humanoids, where the gap between private valuation and commercial evidence is widest. Start with Figure AI. The San Jose company announced that with its Series C it has surpassed $1 billion in committed capital, at a post money valuation of $39 billion. Parkway Venture Capital led the round, with participation from Brookfield Asset Management, NVIDIA, Macquarie Capital, Intel Capital, Align Ventures, Tamarack Global, LG Technology Ventures, Salesforce, T Mobile Ventures, and Qualcomm Ventures. That is a deliberately diversified syndicate: infrastructure money, chip makers, a telecom, an enterprise software company, and a consumer electronics group. A note on timing. The Robot Report article we're drawing from references RoboBusiness 2025 and a December 2024 customer delivery, so check the announcement date against Figure's own materials before you cite it in a model. We include it here because the creator brief places it in this Q3 cycle, and because the terms still frame the humanoid conversation.

Father Zachary

What does the capital buy? Figure says it will build out its Helix embodied intelligence platform and the BotQ high volume production environment it unveiled in March. The company plans to ship 100,000 humanoids over the next four years. It's also building next generation GPU infrastructure with NVIDIA for training and simulation, and collecting more real world data, including human video and multimodal sensory inputs. CEO Brett Adcock said, quote, the team's in place, the robots are built, and the path ahead is clear. Now let me do the scorecard on that. Purchase orders: the article says that in December 2024 the company reported delivering Figure 02 systems to a paying customer. One customer, singular. Deployed unit economics: not disclosed. Gross margin: not disclosed. Balance sheet: over $1 billion committed. Note the word committed. That's not the same as cash received on a single day. And the plan of 100,000 units is a target.

Father Zachary

Here is some interpretive arithmetic, flagged as mine. A $39 billion valuation, divided by 100,000 planned units, is $390,000 of enterprise value per robot in the four year plan. That doesn't tell you what a robot costs or sells for, because the sources don't give a price. It tells you how much the market is paying today for each robot Figure hopes to ship. If the units ship, the valuation is a bet on volume, on manufacturing yield through BotQ, and on a software layer, Helix, that has to keep customers paying after the hardware is delivered. If they don't ship, the valuation is an option on a future. Both can be true at the same time. The job of an investor, and of this show, is to know which one you are holding.

Father Zachary

The same article notes that UBTECH Robotics also announced $1 billion in financing to scale up humanoid production. Two billion dollar financings aimed at manufacturing humanoids tell you capital is being pointed at the factory, not just at the lab. That matters, because manufacturing is where gross margin is made or lost. A software company can raise a billion and spend it on engineers. A hardware company that raises a billion to build a production line is making a commitment to yield, supplier contracts, and inventory. Those are working capital commitments, and as we saw in the RaaS episodes, working capital is where robot companies run out of runway even when demand is real.

Father Zachary

Now the most instructive document of the month. Agility Robotics is preparing to go public through a proposed SPAC merger with Churchill Capital Corp. XI. The Robot Report, in September 2026, got a first look at the company's financials thanks to a new S four filing with the SEC. The headline: Agility reports $1.8 million in revenue, and says it has deployed its Digit humanoid at nine customer sites. It was the most read robotics story of the month on The Robot Report. Alongside the filing, Agility unveiled Digit 5, with more human like legs, upgraded batteries, and a more comprehensive safety architecture. Let me be careful here. The source summary gives us the revenue figure and the nine sites. It does not, in what we have, specify the reporting period, the gross margin, or how many of those nine sites generated that revenue. So any per site math is illustrative only.

Father Zachary

With that caveat, the illustration: 1.8 million dollars across nine sites is about $200,000 per site. If the period is a full year and every site contributed, that's the scale of annual revenue per deployment. If some sites are pilots or came online late, the true recurring figure per site could differ in either direction. The point isn't the quotient. The point is the order of magnitude. A humanoid company with nine deployed sites and seven figures of revenue is a real company with real customers. It is also a very long way from the volume that a multibillion dollar valuation implies. And that is exactly why an S four matters. A filing is a different genre than a demo reel. It has to put revenue, risk factors, and cash on the page. For the anti demo analyst, a document like this is worth ten product videos.

Father Zachary

Then came the public market reality check. Unitree Robotics, the Hangzhou based humanoid maker, listed on August 19. The Robot Report reports that its shares fell nearly 40% in September 2026 from their first day closing price, erasing roughly $20 billion in market value from that point, and about $35 billion from the peak. The same story's headline says shares were down 53% in September from the IPO debut. Those two figures differ, and with the information we have, I can't tell you exactly why. One plausible reading, and this is interpretation, not sourced fact, is that the two numbers use different reference points, such as the first day close versus the intraday peak. Either way the direction is unambiguous. A newly listed humanoid and robot company lost a very large share of its market value within weeks of its debut.

Father Zachary

What should we conclude? Careful. The creator's brief frames this as public investors penalizing hardware commoditization and low unit margins relative to private valuations. That is an interpretation, and the sources we have don't state the cause of the decline. A newly listed stock can fall for reasons of float, lockups, rotating sentiment, or simply an elevated first day price. What we can say is this: private marks and public marks diverged. Figure at $39 billion is a private valuation set by a negotiated round. Unitree is a price discovered daily by an open market. When the two disagree, and when the next humanoid names, like Agility, are lining up to list, the public price is the one that comes with no negotiating leverage. Watch how the Agility SPAC trades, and watch redemption levels, because that will tell you what public investors think about humanoid revenue at the one to ten million dollar scale.

Father Zachary

For a view from the small end of the humanoid spectrum, consider Faraday Future, the Los Angeles company that entered robotics this year after a history as a mobility developer. The Robot Report says it raised $25 million through convertible promissory notes. Combined with $45 million announced in April, that's $70 million over two months. Look at the structure. Of the $25 million, only $12.5 million goes directly into the operating account. The other $12.5 million sits in control accounts, under the investors' control, and is released to the company only after certain conditions are met. The underlying shares are unregistered and not immediately tradable. That is financing with strings, and the strings are the information.

Father Zachary

Now their operations, per the company's own claims. Faraday began selling robots in February 2026 and, as of the end of April, has shipped 68 robots. It hopes to ship 200 in the second quarter and 1,500 by year end. It says it had more than 1,200 nonbinding, paid preorders at launch. Read those words precisely: nonbinding, and paid. First quarter revenue was $512,000, up 62% from $316,000 a year earlier, and nearly equal to the full year 2025 total of $536,000. Ecosystem revenue, meaning skills and software capability packs, was 26% of the total. The company says it has positive single product gross margins. I'll mark that as a company claim; no margin figure is given. So how about the math? From 68 shipped robots to 1,500 by year end is a twenty two fold increase. Whether that happens is the single most useful test of a company like this. It's a clean, falsifiable target. We'll check it in January.

Father Zachary

At the other end of the humanoid market is the consumer. Flourish, a startup spanning San Francisco and Paris, announced plans to sell Flourish One, a wheeled, Raspberry Pi powered humanoid priced at $3,555. The pitch: early adopters can teach it new household chores in under 30 minutes using only a smartphone. The Robot Report ranked it number seven in its September list. It's an interesting product and a clear signal of where the price floor is headed. But the scorecard question is unchanged. Where are the orders, what is the margin at $3,555, and how does a startup fund inventory before revenue? Consumer hardware has a long record of punishing companies on exactly those questions. Interesting is not the same as financeable.

Father Zachary

Let's move down the stack, from the robot to the intelligence inside it. If humanoids are the visible edge of physical AI, foundation models are the invisible core, and that's where some of the largest checks outside defense have landed. Start with Physical Intelligence, the San Francisco company founded in 2024. It raised $600 million in a Series B led by CapitalG, Alphabet's growth fund, with Lux Capital. Bond, Redpoint, and Sequoia Capital also participated, and previous and returning investors include Jeff Bezos, OpenAI, T. Rowe Price, and Thrive Capital. The company has raised $1.1 billion in total and, according to Bloomberg as cited by The Robot Report, is valued at about $5.6 billion. A year before that round, it had raised $400 million. The company says it will use the money to collect more data, make strategic partnerships, and grow its team.

Father Zachary

Why does that matter for capital allocation? Because a language model learns from the web, but a robot has no web to read. It needs real world data. Physical Intelligence's approach, as described by Sacra and summarized by The Robot Report, works like this. A developer streams camera images from any robot into the company's runtime. The system converts that visual stream, along with the robot's movement history, into tokens and feeds them to a transformer model of 3 billion to 5 billion parameters. A user gives a plain language goal, such as make a flat white or pack chocolates into this box. The model takes about 100 milliseconds to predict the next 50 steps, and a hardware abstraction layer turns those tokens into robot specific joint commands, within force and speed limits for safety. That hardware abstraction layer is the strategic piece. It means one model can, in principle, drive many different bodies.

Father Zachary

The company's latest results are worth stating exactly, and as company claims. In February, Physical Intelligence made the code and weights for its Pi zero models open source. Later it announced version 0.6 of its vision language action model. It trained using a method called RECAP, which stands for RL with Experience and Corrections via Advantage conditioned Policies. In plain terms: train a robot by demonstration, coach it with corrections, and let it improve from its own autonomous experience. The company said this doubled throughput on tasks like inserting a filter into an espresso machine, folding previously unseen laundry, and assembling a cardboard box. It also said failure rates decreased over hours of operation and that the method beat imitation learning alone. Doubling throughput is a meaningful claim, because throughput converts directly to the unit economics a customer cares about. But notice what it's measured on: demonstration tasks. The anti demo question is what throughput and uptime look like on a customer's floor, under a customer's contract.

Father Zachary

CapitalG's partners, Jill Greenberg Chase and Manmeet Gujral, wrote that, quote, they are building a single generalist intelligence that manifests in any physical form to solve any real world problem. That's the thesis in one sentence, and it's a big one. The strategic question for an investor is where value accrues if the model is general. If a hardware agnostic model works across many robot bodies, then the model company can become a platform, and the hardware makers can become interchangeable. That is bullish for the model layer and a threat to humanoid makers who bundle their own models. Today's evidence, in the sources we have, doesn't settle it. Other companies are racing in the same direction. Dyna Robotics closed a $120 million Series A for its proprietary foundation model. Skild AI says it's building a general purpose Skild Brain. 1X Technologies uses teleoperation to train its NEO humanoid on household chores. Archetype AI raised $35 million for physical agents, and Foxglove raised $40 million for a data platform for robot developers. Notice the pattern in that last group: even the small rounds are about data.

Father Zachary

Now the opposite strategy: the model with a captive factory. Mind Robotics was spun out in November 2025 by RJ Scaringe, who co founded Rivian. It announced a $400 million raise, bringing its total investment to more than $1 billion. The company obtained $115 million in seed financing in late 2025 and a $500 million Series A in March 2026. Kleiner Perkins led the latest round, with new investors including Meritech Capital, Redpoint Ventures, SV Angel, Incharge Capital, A Star Capital, and Garuda Ventures, and existing investors including Accel, Andreessen Horowitz, Eclipse, Prysm Capital, Bain Capital Ventures, and Greenoaks. Add it up, and a company that is under a year old has raised more than a billion dollars. That's a story about the quality of the syndicate and the clarity of the pitch, not about revenue.

Father Zachary

The pitch is what makes it interesting. Rivian is Mind Robotics' first customer, a key partner, and a shareholder. It provides, in the company's words, a live, high volume manufacturing environment for model training and deployment, which feeds what Mind calls its data flywheel. Compare that to a startup that has to beg for pilot access. Here, the customer is also an owner, and the factory floor is a built in laboratory. The company describes itself as building a full stack platform of foundation models, purpose built robotics, and deployment infrastructure. And here's a detail I'd flag for the anti demo audience. The Robot Report notes that the company's website does not yet have any images of robots. A billion dollars raised, and no robot photographs. I'm not saying that's a flaw. I'm saying it's a useful reminder of what a round at this stage is: a conviction about a team and a distribution advantage, not a product in the field.

Father Zachary

Kleiner Perkins partner Ilya Fushman said, quote, robotics is the ultimate frontier, and that Mind Robotics has unique access to all the ingredients required to make general purpose robotics work in real world manufacturing. Note the phrase unique access. That's the investment thesis, and it is about access to data and deployment, not about a clever algorithm. This echoes a pattern across the sources. The strongest stories pair the intelligence with a place to prove it. Physical Intelligence has partner deployments, such as AgiBot's real world reinforcement learning system in a manufacturing pilot with Longcheer Technology. Mind has Rivian. And in the next case, the place to prove it is a Hyundai plant.

Father Zachary

Boston Dynamics opened its Robotics Metaplant Application Center, called RMAC, at Hyundai Motor Group Metaplant America in Georgia. The Robot Report describes it as a testbed and training center for integrating the Atlas humanoids directly into Hyundai's automotive manufacturing operations. Earlier this year, Boston Dynamics and Google DeepMind announced that DeepMind's Gemini Robotics foundation models will help make Atlas smarter, a reunion nearly a decade after Google sold the robotics company to SoftBank. Put the pieces together: a humanoid maker, a hyperscaler model, and a parent company's live factory. That's an integrated system, and integration is where I'd direct your attention. Notice also that this sits inside Hyundai, and Hyundai is a seller in another deal this month.

Father Zachary

That deal is SoftBank Group's agreement to acquire the Robotics and AI Institute, known as RAI, from Hyundai Motor Group. The Robot Report says the deal had been rumored for a couple of months, is based on multiple sources, and is now under review by the Committee on Foreign Investment in the United States, CFIUS. That review is the detail to track. Cross border robotics transactions now clear a national security lens, and the timeline of that review will tell us something about how Washington treats physical AI research assets. I'll note what we don't have: the price, the terms, and what SoftBank plans to do with RAI. When those arrive, we'll cover them.

Father Zachary

Now the platform layer, where consolidation is quiet but consequential. Qualcomm agreed to acquire PickNik Robotics, the company behind the MoveIt open source manipulation framework. The stated plan is to keep MoveIt open source while streamlining its integration with Qualcomm's Dragonwing platforms, to give developers an easier path from AI models to planning, manipulation, and real time control. In the same month, Arduino, a Qualcomm subsidiary, opened preorders for VENTUNO Q, an edge AI board powered by the Qualcomm Dragonwing IQ8 Series. And Google's Intrinsic released Intrinsic Core, an open source set of ROS compatible capabilities for building robotic applications. Three moves in one month, all pointing the same direction: the chip and cloud giants are competing to own the developer's path from model to motion. My interpretation: the open source label is a distribution strategy. If the framework everyone uses is open but runs best on one company's silicon, the silicon vendor captures the value. The sources don't say that outright, and Qualcomm says it will keep MoveIt open source. But it's what I'd watch.

Father Zachary

Let's step back from the individual deals and look at how the investors themselves explain the surge, because their language reveals what they're underwriting. Crunchbase News spoke with two of them. First, Joe Fath, a partner and head of growth at Eclipse Capital. Eclipse was founded in 2015, and Fath says physical AI is not a new theme for the firm but a core thesis from the start. He argues that, quote, tech barriers are plummeting, experienced talent is pouring in, and market demand is rising. He acknowledges that physical industries remain capital intensive. Historically, the capital required to reach meaningful scale made investors wary. What's changed, in his view, is that AI and enabling technologies are changing how efficiently companies can build and scale. And he says funding is shifting away from experimentation toward companies that can hit production milestones, land customers, and scale efficiently.

Father Zachary

Fath's framing is memorable. Eclipse defines physical AI as intelligence embedded in systems that perceive, reason, and act in the real world, and it generally avoids standalone large language model providers. He describes the firm's focus as investing in the shoulders, rather than the head. The head is the model. The shoulders are what carries the weight: chips, compute, energy, data centers, and the companies that apply AI to build new businesses in the physical world. He expects value to accrue throughout the stack, but believes the strongest moats will belong to companies that vertically integrate and own multiple layers. And his closing line is the one I'd tattoo on every founder's wall: customers value operational efficiency, reliability, and revenue, not technical sophistication alone. That is the anti demo discipline, stated by a growth investor.

Father Zachary

The second voice is Ryan Ziegler, general partner at Edison Partners. He sees the opportunity as broader than the traditional concentration in robotics, humanoids, defense, and foundation models. In his view, physical AI is the convergence of software, hardware, sensors and IoT, and services across many real world applications. What's changing, he says, is AI's ability to process data from those systems at scale and speed to generate useful operational insights, while the underlying hardware gets cheaper and more accessible. He offers a nice example: even our mobile phones now have LIDAR scanners on them, democratizing the ability to map objects and spaces. Edison's interest is in high value, traditionally analog industries, including manufacturing, supply chain, utilities, agriculture, transportation, and government, where physical AI can become mission critical infrastructure. Many of these companies, he says, resemble vertical software businesses, with attractive unit economics, large deal values, and multiyear deployments.

Father Zachary

Here's the part that connects directly to our RaaS work. Ziegler notes that companies are increasingly bundling hardware into recurring or mixed revenue models and moving toward outcome based or usage based pricing. That combination makes the hardware itself, in his words, a distribution mechanism for software and data, a way to create what he calls a data intelligence flywheel. He compares the shift in building costs to what cloud infrastructure did for SaaS. Compute and foundation model capabilities have become more accessible, reusable models and physics based simulation have improved, training data is more plentiful, and sensor and hardware costs have declined. If that thesis holds, then the robot is a loss leader or a low margin delivery vehicle, and the profit is in the recurring layer. Which sounds great, until you remember the RaaS arithmetic: if you deploy the robot on your own balance sheet and collect revenue monthly, you are financing hardware long before the software margin shows up. The flywheel is real only if the company can fund the first turn.

Father Zachary

So the two investor theses, side by side. Fath says own multiple layers and prove reliability. Ziegler says treat hardware as the distribution for a recurring data business. Both describe the same destination: a company that looks like vertical software, with hardware as the on ramp. Where they create tension is the funding path. Vertical integration, which Fath favors, is capital intensive by definition. Hardware as distribution, which Ziegler describes, ties up working capital until the recurring revenue catches up. The sources show both approaches being funded at scale, in Figure's BotQ factory and in Mind Robotics' captive deployment at Rivian. The sources don't tell us which approach produces better returns. That's still an open question, and it is the one I'd press on in every investor meeting.

Father Zachary

Now, the tension the creator brief highlights: consumer hype versus brownfield reality. On one side, capital chasing consumer humanoids, like the $3,555 Flourish One. On the other, heavily capitalized industrial workhorses aimed at manufacturing and logistics shifts that are hard to staff, where plant managers buy verifiable throughput rather than novelty. The Robot Report's reporting from earlier this year offers a data point on the scale of humanoid funding: as much as $4.6 billion was invested in humanoid developers in 2025. Compare that to the $47.4 billion for all physical AI in the first half of 2026 alone. Humanoids are a headline, but in the dollars, they're a minority of the story, and the broader robotics and autonomy complex is where most of the capital sits.

Father Zachary

Let me show you what brownfield capital looks like, using deals from The Robot Report's recent coverage that rarely make a headline. ASI and SoftBank partnered with $225 million to scale brand agnostic autonomous heavy equipment across infrastructure projects, targeting construction labor shortages. Vention raised $110 million for manufacturing automation. ZaiNar raised $100 million for spatial tracking. RLWRLD raised $26 million. The ARM Institute received $90 million for ten projects to modernize military manufacturing. Amazon said it will invest $100 million in a new Indiana manufacturing facility to produce products that power its fulfillment and robotics network across North America. And in January, Amazon acquired Rightbot, a developer of a robot for unstructured truck loading and unloading, a task The Robot Report calls ripe for automation because it is strenuous and happens in all kinds of weather. These are not humanoid stories. They are purchase order stories.

Father Zachary

There's also delivery. Wing Aviation and Walmart plan to expand drone delivery to 150 more Walmart stores over the next year. Wing says that will give more than 40 million Americans access to the service, with a network of over 270 drone delivery locations by 2027. Again, that's the company's and its partner's plan, not realized volume. But it illustrates the pattern: the customer is a retailer with a defined cost per delivery and a clear number to beat. When the buyer has a metric, the sale becomes an engineering and operations problem rather than a belief problem. That's the kind of market where the unit economics can be proven, and investors know it.

Father Zachary

A last observation for this chapter, about how capital is moving across industries this week, which is the question the whole briefing serves. In autonomy, it moves in the billions toward a few category leaders. In defense, toward companies whose customers are governments. In foundation models, toward teams that have data and a place to deploy. In platform software, through acquisitions by silicon and cloud companies. In humanoids, toward a few heavily funded names, with public markets just beginning to price them. And in the industrial base, through targeted investments into manufacturing and logistics. It's one system with many entry points, and the fragmentation of information across those entry points is exactly why a single intelligent hub for market intelligence has strategic value. When the data lives in separate silos, a mega round in one corner and a falling stock in another look unrelated. Put them together and you see the cycle.

Father Zachary

Let's assemble the scorecard for early October 2026, one company at a time, using only what the sources let us say. I'll apply our four tests: purchase orders, deployed unit economics, hardware and software gross margin, and balance sheet endurance. First, Figure AI. Capital: more than $1 billion committed at a $39 billion valuation. Orders: one paying customer reported in December 2024, according to the article, and no order book in our sources. Unit economics and margin: not disclosed. Endurance: strong on paper. Verdict: a high conviction, low disclosure position. The tell to watch is the first audited-scale evidence of BotQ output against the 100,000 unit plan.

Father Zachary

Agility Robotics. Orders: nine customer sites per the S four filing. Revenue: $1.8 million, period not specified in our sources. Margin: not in our sources. Endurance: depends on the SPAC, which means it depends on redemptions and on the market's mood after Unitree. Verdict: the most transparent humanoid company of the month, precisely because it filed. The tell is what the full S four says about gross margin, cash burn, and backlog. If you're researching Agility, read the filing, not the press coverage, and not this podcast.

Father Zachary

Unitree Robotics. It's public, so the market is already grading it. The shares fell nearly 40% in September from the first day close on August 19, with a headline figure of 53%, and roughly $20 billion erased from the first day close and about $35 billion from the peak. Verdict: the clearest evidence that listed robotics names will be priced on fundamentals, and that early enthusiasm can reverse quickly. What I can't tell you from our sources is why. So I'd resist anyone who claims certainty about the cause, including the investment banks and including me.

Father Zachary

Faraday Future. Orders: more than 1,200 nonbinding paid preorders, 68 robots shipped through April. Revenue: $512,000 in the first quarter, with 26% from ecosystem revenue. Margin: the company claims positive single product gross margins, with no figure. Endurance: $70 million raised across two months, half of the latest tranche restricted until conditions are met. Verdict: a transparent target of 1,500 units by year end, which gives us a clean scoreboard. If shipments land near the target, the story improves. If they land at a fraction, the financing structure will matter a great deal.

Father Zachary

Physical Intelligence and Mind Robotics. Both are pre scale in revenue terms, as far as our sources show. Physical Intelligence has $1.1 billion raised and a $5.6 billion valuation, per Bloomberg, with technical claims of doubled throughput on demonstration tasks, open source weights, and partner pilots. Mind Robotics has more than $1 billion raised, a captive first customer in Rivian, and no robot images on its website. Verdict: for these two, the scorecard isn't revenue yet. It's deployment evidence. The question for the next two quarters is whether either announces measurable production results, such as tasks completed, uptime, or throughput under a customer's contract. That is the line between a model and a product.

Father Zachary

Defense and autonomy. Waymo, Anduril, Shield AI, and Saronic make up roughly half of the first half's venture dollars by our arithmetic. The scorecard here looks different because the customers are different. Waymo's $126 billion valuation is a bet on robotaxi scale. The Robot Report even ran a piece asking what it will take to go from 14 cities to 15,000. Anduril's valuation doubled in under a year. The honest caveat is that our sources give valuations and round sizes, not contract backlogs or margins. So the verdict is: capital is concentrated where the buyer has a budget, and that makes the sector sturdier than it looks, but we need the revenue disclosures to confirm it.

Father Zachary

Now the risks, stated plainly. First, concentration. If one third of the half's dollars came from a single round, the headline sum is fragile to a single decision by a single investor group. Second, the exit gap. Robotics proper has had fewer exits than defense and space. The Mentee acquisition by Mobileye, at roughly $900 million, is one of the few, and the first humanoid listings are now being tested by the market. Third, financing structures. Convertible notes with restricted tranches, like Faraday's, and SPAC mergers, like Agility's, are not signs of weakness by themselves, but they're signals about the cost and availability of capital at the smaller end. Fourth, regulatory friction. CFIUS review of the SoftBank and RAI transaction suggests that cross border deals will not all clear quickly.

Father Zachary

And what would make us more constructive? Four things. One: disclosed purchase orders and backlog, from any humanoid or general purpose robot maker, in dollars rather than in units announced. Two: gross margin figures that separate hardware from software and services, so we can test the hardware as distribution thesis against the numbers. Three: funded deployments where customers pay on usage or outcomes and the vendor can finance the hardware without diluting the equity. Four: a successful robotics listing that holds its value past the first quarter. If those arrive, the 2026 surge starts to look like the early part of a durable cycle. If they don't, it looks like a mega round cycle that outran its evidence.

Father Zachary

Here's the calendar to keep. Agility's SPAC process and the full S four. The Faraday Future shipment pace against its 1,500 unit year end target. The CFIUS review of SoftBank and RAI. Any financial disclosure from Figure about BotQ output. The next monthly total after August's $4.9 billion in robotics investments, so we can see whether the run rate holds without a Waymo sized round. And RoboBusiness in Santa Clara, where The Robot Report's parent is celebrating twenty years of the event, with investors such as F Prime Capital scheduled to speak on the state of robotics investments. Those are the checkpoints where claims turn into evidence.

Father Zachary

Let me close with the thesis, in one paragraph. The first nine months of 2026 delivered the largest influx of capital into physical AI on record: $47.4 billion in the first half, more than the three years from 2022 to 2024 combined, with about half concentrated in four autonomy and defense rounds. Humanoids and foundation models absorbed billion dollar checks, and the platform layer consolidated through acquisitions by Qualcomm and SoftBank. Then September brought the first honest public prices: a thin but real set of filings from Agility, and a steep slide for Unitree. The capital is real. The evidence is catching up. And the investors who do best from here will be the ones who ask not what the robot can do in a video, but what the customer pays for, what it costs to deliver, and how long the company can last while it finds out.

Father Zachary

That's the briefing. This is ROBOT CAPITAL, brought to you by IRHUB and KMKT. The sources for today's episode include Crunchbase News, the Wall Street Journal, and The Robot Report, and the company claims and interpretations I've flagged are exactly that. Nothing here is investment advice. Follow the companies. Follow the technology. Follow the capital. And until next time, don't follow the demo.