Where the Money Goes: Physical AI and the Q3 Reindustrialization Wave

Zach Martin

Welcome to Robot Capital, on KMKT IR Hub Radio. I'm Zach Martin, it's Monday, October 12, 2026, and there's a lot of money to follow this week. Let me start with a sentence that sounds like a contradiction. Global venture funding in the third quarter totaled $159 billion, and that was the lowest quarter of this year so far. Yet it was still higher than every other quarter going back to the second quarter of 2022. That's Crunchbase data, as of October 2. So if someone tells you funding fell in Q3, they're right. If someone tells you we're in a historic funding boom, they're also right. Both statements are the same number, viewed from two sides. Today I want to walk through that number, because buried inside it is the story of where capital goes when software isn't the only place to put it. It's going into power, silicon, satellites, factories, and machines that actually move.

Zach Martin

If you caught Episode 8, we talked about the $47 billion physical AI wave and the hardware rebound. That was the first half of the year, and we looked at the early mega rounds for Waymo, Anduril, Shield AI, and Saronic. This episode is the sequel, but you don't need that one to follow along. We're doing three things. First, the third quarter post mortem. Second, the deals that hit the wires this week. Third, and this is the part I find most interesting, a look at how the money is splitting apart. The brains of robots are being financed separately from the bodies, and the work of keeping the electric grid running is starting to look like a robotics market. Let's start with the scoreboard.

Zach Martin

Here are the headline figures from Crunchbase. Close to 6,000 startups were funded in Q3, for that $159 billion. Add up the first three quarters of 2026 and you get $679 billion in global venture funding, which Crunchbase says is the highest amount in the first three quarters of any year on record. Compared with the second quarter, Q3 was down 25 percent from $212 billion. Compared with the same quarter last year, it was up 53 percent from $104 billion. Now, why the dip from Q2? Crunchbase's Joanna Glasner, writing about investor activity, tied the decline to the absence of new record megarounds for the AI leaders. In other words, it's not that the deal makers went home. The most active investors, names like Andreessen Horowitz, Insight Partners, and Sequoia Capital, mostly did more deals than in the prior quarter. The dollars per deal at the very top just didn't set new records.

Zach Martin

And the very top is where the story lives. An all time record 27 companies raised rounds of a billion dollars or more in Q3. The previous record was 16 in the second quarter, and the first quarter had 14. So that count nearly doubled in two quarters. Around a third of all global venture capital raised in the quarter went to those 27 companies. Think about the proportions there. Close to 6,000 startups were funded, and about a third of the money went to 27 of them, which is well under one percent of the companies. No single company raised in the tens of billions. But eight companies raised $3 billion or more, and five of those eight were founded within the past four years. The biggest checks went to Databricks and Safe Superintelligence, $5 billion each. Then came Crusoe, Moonshot AI, Mistral AI, Nscale, The Boring Company, and Kling AI, each north of $3 billion.

Zach Martin

So where does physical AI fit in a quarter dominated by frontier labs and data centers? Start with AI overall. AI driven startups across the stack raised $102 billion, which is 64 percent of global venture, down from the previous two quarters but still 14 percentage points above the share in Q3 of last year. The United States took $91 billion, about 57 percent of the global total, and the San Francisco Bay Area alone took 24 percent. Half of the capital invested in Q3 went to companies founded since 2022. That last stat is worth sitting with. The money is chasing very young companies. Now the physical piece. Crunchbase says that in Q3 the aerospace, robotics, data center, semiconductor, and energy sectors each raised $10 billion or more. Five separate sectors, each at ten billion or above, in one quarter.

Zach Martin

Here's how I read that list, and I'll flag that this is my interpretation, not Crunchbase's. Look at the sectors as a chain. A robot, or a self driving truck, or a data center full of GPUs, needs chips to think, power to run, and a factory or a launch pad or a fab to get built. Data centers, semiconductors, and energy are the inputs. Robotics and aerospace are some of the outputs. Capital is moving up and down that whole chain at once. Crunchbase's own summary is that venture funding has expanded beyond software into physical infrastructure and robotics, and it uses the phrase reindustrializing economies. That's their framing. What I can say from the numbers is that one quarter had five physical sectors each clearing ten billion dollars, and that wasn't true of venture for most of the last decade.

Zach Martin

Let me anchor that with the half year numbers, because definitions matter and I want to be careful. Crunchbase counts physical AI as robotics, autonomous vehicles, aerospace, drones, industrial automation, and sensors. By that definition, the first half of 2026 saw $47.4 billion across 521 deals. That's almost four times the $12 billion across 470 deals in the second half of 2025, and nearly 80 percent above the $26.4 billion in the first half of last year. And for scale, all of 2022 through 2024 combined, three full years, totaled $41.9 billion. So the first six months of this year beat the prior three years put together. A Forbes piece by Sabbir Rangwala puts the first half at roughly $50 billion across roughly 500 deals. Same ballpark, slightly different bucket. PitchBook, using its own robotics and physical AI category, counted a record $16.3 billion across 492 deals in the first quarter alone. These are different taxonomies, so I wouldn't add them together. I'd treat them as three independent instruments all pointing the same direction.

Zach Martin

One reason the first half number is so large is concentration. Waymo's $16 billion Series D in February, at a $126 billion valuation, accounted for nearly a third of all the physical AI venture dollars in that half. Alphabet, Dragoneer, DST Global, and Sequoia co led. Then Anduril raised another $5 billion in May at a $61 billion valuation, double the $30.5 billion it was valued at less than a year earlier. Shield AI raised a $2 billion Series G in March at a $12.7 billion valuation, and Saronic, which builds autonomous sea vessels, raised $1.75 billion in Series D at $9.25 billion, more than double its prior level. If you pulled those handful of deals out, the story would be quieter. But PitchBook's point for Q1 is that the spike is the surface story. The deeper one, in its words, is that robotics is crossing from experimentation into deployment.

Zach Martin

Now exits, because money in is one thing and money back out is another. The first half had SpaceX raising $75 billion in its June IPO at a $1.77 trillion valuation. HawkEye 360 raised $416 million and the drone maker Aevex raised $320 million in their listings. On the acquisition side, Mobileye paid roughly $900 million for the Tel Aviv humanoid startup Mentee Robotics, and Mobileye explicitly tied that deal to its push into physical AI. Crunchbase notes that exit activity has been more concentrated in aerospace, defense, and drones than in robotics proper. Then Q3. China dominated the IPO side. ChangXin Memory Technologies, a DRAM chip developer, raised $8.6 billion at an $85.5 billion value, and its stock surged more than 500 percent on its debut. The Hangzhou humanoid maker Unitree Robotics went public, raising $905 million at a $9 billion valuation. Crunchbase says semiconductor and robotics listings surged on their first day of trading.

Zach Martin

The acquisition headlines in Q3 are where it gets interesting for our beat. AMD announced its intent to acquire Fei Fei Li's World Labs for $8.2 billion. World Labs builds models for physical intelligence, meaning models that help machines make sense of space and objects. Nvidia made a bid to acquire Hugging Face for $12.9 billion. Crunchbase notes that semiconductor companies were the largest acquirers in the quarter. I'll offer a cautious reading. The companies that sell the picks and shovels for AI are buying the software that teaches machines about the physical world. I can't tell you the strategy inside those boardrooms, and the sources don't say. But the pattern of chipmakers bidding for models of physical intelligence is a fact worth keeping on your radar.

Zach Martin

Who is writing the checks? Forbes contributor Sabbir Rangwala points out that Wall Street is getting in on the action, and he lists JP Morgan, SoftBank, Goldman Sachs, KKR, and Blackstone. And Crunchbase's investor rankings show who's most active. Y Combinator led with 45 post seed deals and participated in at least 221 seed rounds in the quarter. Valor Equity Partners and Atreides Management tied for the most capital put to work as lead investors, $7.7 billion each in led or co led deals, with overlap across at least six rounds, including Crusoe's $3.9 billion Series F. Andreessen Horowitz and Nvidia came next at $6.5 billion and $6.3 billion, and Nvidia's standout was that $5 billion round for Safe Superintelligence. The takeaway, in Crunchbase's words, is that the same familiar names top the rankings, but a few firms, Atreides and Valor in particular, are significantly scaling up.

Zach Martin

Let me close this scoreboard with two investor voices who explain the why, from that Crunchbase physical AI piece. Ryan Ziegler, a general partner at Edison Partners, says the costs to build these companies have come down, and compares it to what cloud infrastructure did for software as a service. He describes hardware as, and I'm quoting, the distribution model for creating a data intelligence flywheel. In plain English, you sell or lease a machine, and the machine collects the data that makes your software smarter. Joe Fath, a partner at Eclipse Capital, says funding is shifting away from experimentation and toward companies that can hit production milestones, land customers, and scale efficiently. Eclipse talks about investing on the shoulders rather than the head, meaning the chips, compute, energy, and data centers, plus the companies applying AI in the physical world, but generally not standalone language model providers. Keep that shoulders and head idea in your back pocket, because the deals this week fit it almost perfectly.

Zach Martin

Now to this week, which Crunchbase tracked as October 3 through 9. Joanna Glasner's weekly list of the biggest U.S. rounds opens with a number that tells you how wide the net has gotten. The biggest round of the week, $2 billion, went to Axiom Solutions International, a private cloud and power infrastructure company spun out of Flex, the manufacturing services provider. Austin based Axiom drew General Catalyst and Koch Equity at an initial enterprise value of $37.5 billion. Flex says it plans to separate Axiom into an independent publicly traded company early next year. So that's a spinout, a strategic investment, and a planned public listing all in one transaction. And the company's pitch is the AI buildout, the cloud and the power that has to sit behind it.

Zach Martin

I want to add a footnote here, because headline numbers can mislead. Crunchbase describes this as General Catalyst and Koch Equity providing a strategic investment by purchasing shares of Axiom from Flex. That wording suggests the structure involves buying stock from the parent. The source doesn't break out how much of the two billion dollars went onto Axiom's own balance sheet versus to Flex, and I won't guess. What we can say is that two serious investors paid for a stake at a $37.5 billion starting valuation in a company whose business is power and private cloud. For a manufacturing services company, which is a sector most people think of as contract assembly of other people's products, that's a striking valuation for one piece of it.

Zach Martin

Second on the list is TypeSafe AI, developer of a model called Jev, which has a large and growing early following. It raised $870 million, with Andreessen Horowitz as the largest investor and Sequoia Capital also participating, at a reported $7.5 billion valuation for the San Francisco company. Per Crunchbase, that came from a Bloomberg report, so treat the valuation as reported rather than confirmed by the company. Now, I'll be honest with you. TypeSafe isn't a robotics company. It's a foundational AI model developer. I include it because it's the same wave. The model layer and the machine layer are being funded by overlapping investors in the same weeks, and a lot of the thesis for robot intelligence assumes that general purpose models keep getting better and cheaper to run. We'll come back to that tension when we talk about robot brains.

Zach Martin

The deal I want to spend real time on is smaller in dollars and bigger in what it implies. Atomic Machines, based in Emeryville, California, came out of stealth this week and said it has raised $250 million to date from a long list of investors. Note that phrasing, to date, which means that's a cumulative figure, not necessarily one new round. What do they do? Crunchbase describes it as a startup working on AI native digital manufacturing systems to build micro machines directly from code. Let me say that slowly. The idea is that you write software that describes a tiny machine, and the manufacturing system builds the physical thing from that description. The sources give us no customer list, no product specifics, no revenue, so I can't tell you whether it works at scale. What I can tell you is that investors put a quarter of a billion dollars behind the premise that manufacturing itself can become a software problem.

Zach Martin

Pair that with another deal from the same list. Vinci, based in Palo Alto, closed $250 million in Series B funding at a $1.5 billion valuation. Advent, Temasek, and Xora led. Vinci develops an AI computational platform for hardware engineers to design products. So on one end of the pipeline you have software that helps engineers design physical things, and on the other you have a company building a manufacturing system to produce physical things from code. Now, to be clear, these are two separate companies and nothing in the sources links them. But sit them next to each other and you can see why the phrase physical AI keeps stretching. It's not only about a robot arm with a better camera. It's about shortening the loop between an idea in a computer and an object in the world. In the Forbes piece, Linse Capital's seed investments make the same point with names like Amca, which designs and builds critical aerospace and defense components using physics and geometry based systems for concurrent design, prototyping, testing, and production.

Zach Martin

Now the energy side of the ledger. Voltus, a San Francisco provider of a distributed energy platform aimed at bolstering grid flexibility, picked up $225 million in Series D financing, led by Generation Investment Management, Activate Capital, and Vitol. Type One Energy, in Knoxville, Tennessee, a developer of fusion energy technology, secured $200 million in Series B financing led by Breakthrough Energy Ventures and Clutterbuck Capital. And in quantum computing, Oratomic of Pasadena raised $475 million led by Arch Venture Partners, Spark Capital, Khosla Ventures, Index Ventures, General Catalyst, and Bezos Expeditions, after a $300 million financing in July. Why am I bundling them? Because that same Crunchbase list described the week as featuring AI infrastructure providers and, in the article's words, investors leaving few niches unfunded. And the Forbes piece I'll get to later makes the link explicit: fusion power matters to this crowd because it's seen as critical for powering the energy hungry data centers that support AI. That's the author's framing about Thea Energy, a Linse portfolio company, but the logic carries.

Zach Martin

Let me back up one week, to the list that ran October 2 and covered September 26 through October 2, because several physical AI deals landed in that window. Etched closed a $300 million Series C at a $10.3 billion valuation, led by Sequoia. That price doubled the $5 billion valuation Stripes set earlier this year. Now here's the careful part. A deal teardown quoted by the data aggregator Signalbase says the step up is anchored to $1 billion in booked contract orders, yet the company's most recent disclosure put its first inference chip in customer testing with no named buyers and no recognized revenue. So booked orders are a promise, and recognized revenue is cash. Both of those can be true at once, and the gap between them is where a ten billion dollar valuation either earns its keep or doesn't. I'm reporting that qualification because it's the kind of detail the headline skips.

Zach Martin

Also in that window, General Intuition secured $220 million at a $6.2 billion valuation, with Valor Equity Partners and Atreides Management as the lead backers. The New York company develops AI models trained on videos and virtual worlds to develop skills for physical environments. That's the idea that a machine can learn how the physical world behaves by watching enough footage and simulating enough worlds. SiMa.ai, from San Jose, picked up $150 million in Series C financing at a $1.45 billion valuation, led by Fidelity and Amplify. It develops a software platform for scaling physical AI applications, and the brief describes it as edge physical AI processing, meaning the computing happens on the machine rather than in a faraway data center. And Varda Space Industries raised a $251 million Series D. Varda processes materials in orbit and returns them to Earth, building the in orbit production equipment and the reentry capsules. Signalbase's list shows that round twice under two entity names, so don't count it double.

Zach Martin

I want to make a quick point about reading these weekly lists, because it's a trap I'd hate for you to fall into. Look at GMI Cloud, a provider of GPU cloud infrastructure for AI teams. Signalbase lists it at roughly $668 million for the week ending October 4, as the biggest U.S. deal it tracked. Crunchbase lists it at $223 million in Series B funding, with Archiv leading and Nvidia participating, plus $445 million in debt financing. Add those two and you get Signalbase's number. So the same company, the same week, shows up as $223 million or $668 million depending on whether you count debt. Same with Homeward, which closed $120 million in equity and another $330 million in debt. Debt matters a lot for physical infrastructure, because buildings and machines and chips can be financed against assets. But a venture round and a loan are different animals, and when you hear a giant number, ask which one it is.

Zach Martin

And remember the weekly lists are a mix. The October 2 list was topped by Instinct, a personal AI assistant startup, with a $1 billion Series C at a $10 billion valuation. EliseAI, which deploys AI for the home rental industry, raised $350 million. Armadin, an AI cybersecurity startup, raised $255.5 million. Signalbase counted 41 verified U.S. rounds that week totaling roughly $3.7 billion, with Y Combinator the most active investor at four deals. So physical AI isn't swallowing the venture market. It's a growing slice of a market still tilted heavily toward software. That's why the Q3 statistic matters. The physical sectors each cleared ten billion dollars, and the whole quarter was $159 billion. They're real, they're growing, and they're still a minority of the total.

Zach Martin

So what's the common thread this week? I'd say it's the stuff underneath the robot. Power at Axiom and Voltus and Type One. Silicon at Etched and SiMa.ai. Design software at Vinci. Manufacturing from code at Atomic Machines. Models of the physical world at General Intuition. Orbital factories at Varda. Not one of those is a humanoid doing laundry. But every one of them is a bet that the physical economy gets reshaped by cheaper intelligence. When we turn to the machines themselves, there's a different and sharper question. If the intelligence gets separated from the machine, who owns the value? Let's get into that.

Zach Martin

Here's a football analogy that helps me with this, and I'll call it an analogy so nobody thinks it's data. Imagine a quarterback who can run any offense with any group of receivers, any offensive line, in any stadium. That's the dream of a horizontal robot brain, one set of software that can drive many different robot bodies. Now imagine the opposite, a team built entirely around one system, with specific players for specific plays. That's the vertically integrated company, hardware and software built together. Venture capital is now funding both models, at a scale that would have sounded absurd a few years ago, and the New Market Pitch data shows just how fast the leaderboard has changed.

Zach Martin

Start with the numbers. New Market Pitch, in a report updated September 23, says Skild AI now appears to sit at the top of the robotics startup funding market, with roughly $2.2 billion in reconstructed cumulative financing. That includes a $1.4 billion Series C announced in January 2026. Notice the word reconstructed. These are pieced together from public reports rather than audited totals. Behind Skild is a new billion dollar group: Figure, which has raised more than $1.7 billion once its Series C of more than $1 billion is combined with earlier rounds, and the brief puts its valuation at $39 billion. NEURA Robotics, which announced a Series C of up to $1.4 billion in June 2026, and which the brief says is backed by Tether, Qualcomm, Amazon, Nvidia, Bosch, Schaeffler, and the European Investment Bank. Physical Intelligence, which TechCrunch reported had raised just over $1 billion. And Apptronik, with $963 million raised, including a $520 million extension in February that the brief says drew Mercedes Benz, Google DeepMind, John Deere, and the Qatar Investment Authority.

Zach Martin

Look at how young those companies are. Skild AI was founded in 2023, Figure in 2022, and Physical Intelligence in 2024. Yet each is in or around the billion dollar funding class. The previous generation of robotics took a decade or more to raise that kind of money. And the sizes of the individual rounds are what's new. Skild and NEURA announced rounds of up to $1.4 billion, Figure raised more than a billion, and Apptronik added $520 million in a single extension. New Market Pitch's phrase is that these are financing sizes that used to be almost unheard of in robotics. For context on the whole category, it estimates roughly $15 billion went into robotics startups globally in 2025, and that the running 2026 total had already reached about $18.8 billion by late June, above the previous annual record.

Zach Martin

Now the idea I find most important in this whole briefing. New Market Pitch's summary says robot intelligence is increasingly being financed separately from robot hardware. Skild AI, FieldAI, and Physical Intelligence are pitching general purpose control systems that could work across multiple robot bodies and industries. FieldAI, per the brief, has raised $405 million across two rounds. Why does that matter? Because it changes the size of the market an investor can imagine. A company that builds one kind of robot for one kind of job has a market equal to that job. A company that sells the brain to anyone who builds a robot has, in theory, a market equal to all of them. That's what New Market Pitch means by a much broader theoretical market. And the word theoretical is carrying real weight there. It's a thesis, not yet a proven outcome.

Zach Martin

The brief gives one data point toward proof, and I want to handle it carefully. Skild AI is reported to have reached $100 million in annual recurring revenue within 10 months of its initial commercial deployment. If that's accurate, that is an extraordinary pace for a company whose product is a robot control model. But notice what we don't know, and the brief flags it as a critical uncertainty. How much of that hundred million is sticky, multi year software contracts, and how much is upfront enterprise pilot commitments? Those are very different things. A pilot is a company saying let's try this on a few machines. A multi year contract is a company saying we're building our operations around this. Annual recurring revenue is supposed to mean the second thing, but in a young market the label can stretch. I'm reporting the number as reported, and I'm telling you the question that matters is the one behind it.

Zach Martin

Another caution on Physical Intelligence. Bloomberg reported discussions for another roughly $1 billion round for the company. New Market Pitch says it would keep a proposed round separate until completion is clearly documented. I'll do the same. Just over a billion raised is the supported figure. Talks of another billion are talks. This matters because in a hot market, rumored rounds get repeated until they sound like facts, and then valuations anchor on them. So when you see a leaderboard of robotics funding, check which numbers are closed and which are in discussion.

Zach Martin

Now the humanoids, because that's where the eye popping growth is. Dealroom's robotics sector page ranks humanoid as the hottest sub segment, with VC growth up 2,635 percent, and 18 unicorns worth $73.5 billion in enterprise value. Per the brief, humanoid funding has hit about $11.0 billion in 2026 alone. And here's the geographic breakdown from Dealroom's 2026 humanoid data through July 22. China accounts for 63 percent of global humanoid venture investment, $6.9 billion. Europe is 15 percent, $1.6 billion. The United States is 12 percent, $1.3 billion. Let me repeat that, because many listeners assume the U.S. leads. In humanoid venture funding this year, through late July, China is roughly two thirds of the total, and the U.S. is about an eighth.

Zach Martin

Why China? The brief attributes it to dense regional component supply chains, things like high torque actuators, harmonic drives, and tactile sensor arrays, along with national manufacturing clusters. That's the explanation the briefing offers, and it's an interpretation rather than a measured cause, but it's plausible on its face. If the parts are made nearby, iterating on a robot is faster. New Market Pitch adds that China has become much harder to treat as a secondary robotics market, with Chinese companies already holding important positions in warehouse, service, and industrial robotics. And remember what happened in Q3. Unitree Robotics, the Hangzhou humanoid maker, went public raising $905 million at a $9 billion valuation. So a Chinese humanoid maker has already crossed into public markets, while many of its Western peers are still raising private rounds.

Zach Martin

Now the historical baseline, because it keeps the humanoid hype honest. In the New Market Pitch dataset of 95 companies, warehouse and logistics robotics represent roughly $6.6 billion of cumulative funding, and the brief counts 27 companies there, including Geek+, Symbotic, Exotec, and Locus Robotics. Surgical robotics holds about $5.7 billion across 17 companies, and CMR Surgical leads that dataset at $1.2 billion, with its Versius robot used in more than 30,000 procedures. Explicit humanoid companies in that historical dataset total only around $1.8 billion. Yet in 2026 alone, the brief says humanoids have drawn $11.0 billion. So the pool of money accumulated over the past decade sits in specialized machines that already work, and the fastest new money is flowing to the general purpose bets. New Market Pitch describes it as a market splitting into two funding eras at once.

Zach Martin

Is the old guard still getting funded? Yes, quietly. Locus Robotics added a $42 million Series G in August. Vecna Robotics raised a $31 million Series D in September. Standard Bots raised a $200 million Series C in June, with General Catalyst and the Amazon Industrial Innovation Fund among its investors. Pudu Robotics, which makes commercial service and delivery robots, raised $150 million in April. These aren't ten figure headlines, but they're the businesses with customers today. A warehouse operator buying autonomous forklifts or pickers isn't waiting for a humanoid to show up. And New Market Pitch notes that industrial robotics is producing new billion dollar valuations as AI makes factory automation easier to deploy.

Zach Martin

Let me also flag a caveat that New Market Pitch itself raises about the leaderboard. Cumulative funding measures financial firepower, not who has won. The ranking mixes private startups, public companies, acquired businesses, and companies that later shut down, and some totals include IPOs, PIPEs, at the market offerings, or de SPAC capital rather than venture rounds alone. So when I say Skild has raised $2.2 billion and CMR Surgical $1.2 billion, I'm not telling you who is better, or who is more valuable. I'm telling you who has had the most capital put in. In sports terms, that's a payroll ranking, not a standings table. Some of the biggest payrolls have missed the playoffs, and we'll get to those teams in the last section.

Zach Martin

So the central strategic question is this. If robot intelligence is a separate layer, does it become a durable business, or a commodity? The brief frames it as an open question: will the pure play robot intelligence platforms, Skild, FieldAI, and Physical Intelligence, become commoditized horizontal layers captured by hyperscalers, or will vertically integrated hardware and software players establish deeper moats? Eclipse's Joe Fath, from that Crunchbase piece, leans toward the second view. He says the strongest moats will belong to companies that vertically integrate and own multiple layers. He also says customers value, and I'm quoting, operational efficiency, reliability, and revenue, not technical sophistication alone. That's one investor's expectation, not a verdict. But notice that it cuts against the horizontal brain thesis, and yet Eclipse is clearly investing in this space. So even the people who like the sector disagree about where the value pools.

Zach Martin

There's a hint of a middle path in a theme the Forbes piece raises. Hardware as a distribution mechanism. When a machine in the field is generating data, the company that owns the machine, the sensors, and the software has a flywheel the pure software company lacks. At the same time, an OEM agnostic software company, which works across brands, can argue the reverse, that nobody wants a closed silo. Both arguments can be right in different industries. And that leads to my favorite part of this briefing, which is what happens when the machines leave the lab and meet the customer's real yard, real fleet, and real power grid. That's next.

Zach Martin

Sabbir Rangwala, in that October 5 Forbes piece, starts with a framing I want to unpack. About three decades ago, the Internet of Things connected static machines to computers. A pump, a meter, a turbine could report its status to a server. Rangwala uses the term AoT, Autonomy of Things, a registered trademark of Patience Consulting, so it's his vocabulary and not an industry standard. In his framing, AoT connects static and moving things, sensors, and localization data to create autonomous movement. The difference between IoT and AoT is the difference between a machine that tells you it's broken and a machine that drives itself to the problem. The brief adds the technical ingredients: multispectral cameras, solid state 3D LiDAR, compute, and localization all working together to execute physical motion.

Zach Martin

He sorts the environments into three tiers, and I find this the most useful mental map in the whole stack. Uncontrolled environments are things like autonomous cars in dense urban settings. Anything can happen, a pedestrian, a cyclist, a ball rolling into the street. Semi controlled environments include autonomous trucking on highways, blue collar vehicles in construction, agriculture and mining, drones in regulated airspace, and robots or drones monitoring infrastructure. Controlled environments are robots in industrial settings and campuses, rail and shipping yards, and warehouse logistics. The more controlled the place, the sooner the robot can earn money. And Rangwala notes the operating conditions are typically harsh: weather, dust, vibration, and shock. A demo in a clean lab is not that. His list of what drives all this is capital efficiency, productivity, quality of life, safety, and what he calls the acute shortage of skilled human labor in various physical industries.

Zach Martin

Now the history, which explains why venture capital stayed away for so long. Rangwala says the autonomy revolution was driven largely by big corporations. Caterpillar and John Deere have invested in partial or full autonomy for at least three decades. The DARPA Grand Challenge of 2006 spurred the driverless car movement, initially pioneered by Google, which became Waymo, then Baidu, Uber, Tesla, and the large automotive OEMs. And his blunt line is that physical AI is tough. It takes patience, innovation, data, and extensive testing before it can generate revenue, and traditional venture capital stayed away from it. Think about that against the numbers we just covered. The thing that scared investors off for decades is exactly the thing they're now underwriting at ten and eleven figure valuations. What changed? Ziegler and Fath both said costs fell and tools improved. Rangwala points to specialized deep tech venture firms backing alumni of tech companies, OEMs, and universities, in companies like Wayve for driverless cars, Aurora, Waabi, and Gatik for trucks, and BrightAI for infrastructure monitoring.

Zach Martin

Here is what Rangwala calls a key emerging trend, and it's one I think is underreported. Companies like ASI and AIM are building OEM agnostic physical AI stacks. The brief identifies ASI as Autonomous Solutions Inc. Let me explain why that's a big deal with an example. Picture a mine, or a large construction contractor, or a big farm. Their fleet isn't one brand. They own machines from many manufacturers, say Caterpillar, John Deere, Komatsu, or Volvo, bought over years. If each brand offers its own autonomy system, then each brand's machines work in their own silo, and the customer ends up running several incompatible systems, which is piecemeal. An OEM agnostic company writes one autonomy and fleet management layer that bolts onto machines from different manufacturers. Rangwala says that allows the operator to run the fleet autonomously, rather than piece meal, enabling a coherent, productive, and efficient process.

Zach Martin

Think about why that's an investable idea. The customer owns the problem, which is a mixed fleet. The incumbent manufacturers each have an incentive to sell a closed system. A third party that works across brands lowers procurement barriers, according to the brief, because the customer doesn't have to rip out and replace their existing iron. That's the logic as the briefing lays it out. I'd add that it's a claim about how the market could develop, and I haven't seen revenue figures for ASI or AIM in the supplied sources. But it connects directly to the horizontal brain thesis we just discussed. It's the same architecture, intelligence separated from the body, applied to heavy equipment instead of humanoids. And the heavy equipment case may be easier, because the environments are semi controlled and the customer's economics are clear.

Zach Martin

Rangwala also names supply chain resiliency and sovereignty as another driver of venture investment in physical AI. This is where Linse Capital comes in. Linse has backed deep tech and physical AI companies for two decades, with investments in Skydio, Waabi, and Wayve. Historically it did late rounds, taking controlling positions with checks of about $100 million per deal, roughly $1.7 billion in total. Now it has launched an early stage fund, Linse Ignition, which has raised and deployed $75 million since its launch in 2025, with another $75 million in co investment vehicles, across 14 startups from an initial funnel of 400 opportunities. Bastiaan Janmaat, a general partner at Linse, said they can now put their money where their mouth is and deploy early checks, working with founders from those early stages. Notice that a firm known for big late stage checks is walking backward down the stage ladder. That tells you where the scarce opportunities are perceived to be.

Zach Martin

The Linse seed portfolio reads like a tour of the physical AI supply chain. Amca designs and manufactures critical components for planes, military vehicles, and core infrastructure. Varda, as we said, processes materials in orbit. Fab2 builds compact modular chip fabrication facilities and the machinery inside them. Lace Lithography is developing what it calls beyond EUV atomic scale lithography systems to enable quantum technologies and physical AI. Mythic is building analog compute in memory architecture that Rangwala says delivers up to 100 times greater efficiency for edge computing and AI. And Thea Energy is pursuing a faster, simpler approach to commercializing fusion. Notice what isn't on that list. There are no humanoids. It's the plumbing: chips, power, components, and compute at the edge. That's the shoulders, in Eclipse's terminology.

Zach Martin

Now the piece I most want you to hear, because it's a case where a venture thesis is a specific, testable claim about work. SET Ventures in Amsterdam and Future Energy Ventures, known as FEV, in Berlin, recently released a white paper on how robotics and physical AI can support the energy grid. Some background. SET was founded in 2007, with Fund III at €100 million and Fund IV at €200 million active. It invests early. FEV was founded in 2016 as a spin off of Innogy Ventures, which was backed by the German utility E.ON, and has about $235 million in funding across energy. It invests later. According to Forbes, the white paper researched and mapped 600 companies. That's a real landscape study, not a press release, and the conclusions are specific.

Zach Martin

The white paper ranks investment priorities, and the ordering is the story. First, operations and orchestration. As robots deploy into the field, the layer that coordinates them rises in value, and the stickiness comes from logic embedded in each company's specific workflows. Second, physical work. Customers increasingly want robots to do physical work, like repairing a piece of equipment. Companies with proprietary technology, recurring robotics as a service models, and clear operational advantages are an investment priority. Third, inspection and monitoring. That has, in the paper's telling, progressed dramatically with robots and drones, so there are fewer high quality investments left, though it remains an area of selective focus. And fourth, asset intelligence software, which is not an area of focus, because advances in AI have eliminated barriers to entry and increased commoditization of the base asset intelligence layer.

Zach Martin

Let me translate that into a plain picture. Say there's a wind farm. Years ago, a company might have sold you software that told you which turbines were likely to fail. Then drones came along and flew around capturing images of the blades. Those two things, the dashboard and the drone, are now crowded and cheaper. The white paper's investors say the next frontier is the robot that doesn't just flag the crack but fixes it. Repairs a valve on an oil pipeline. Repairs and reattaches electric power cables. The brief points to a robot from a company called Cosmic installing solar panels, actual physical work in solar panel assembly. And the paper notes this is already occurring in agriculture, construction, and mining, as robots, sensors, compute, and AI mature and fall in cost.

Zach Martin

The two investors quoted in Forbes put it in memorable ways. Austin Wood of SET Ventures said, and I'm quoting, the next wave of physical AI in energy will be won on integration. The winners will take proven advances in autonomy, batteries, and actuators and engineer fit to task systems that are reliable in the field and profitable from day one. He added that with hardware and compute costs at historic lows, the real challenge is finding the environments where a purpose built system can beat human labor on cost or productivity, then optimizing performance relentlessly. And Moritz Jungmann of FEV said physical AI in energy is about machines that act, not just observe. Drones have gotten very good at capturing data, but the next generation will interact with what they find, so robots might actually fix something, not just flag repairs. I like that framing because it's a falsifiable test. Either a purpose built machine beats a human crew on cost, or it doesn't.

Zach Martin

Put this next to this week's deals. Voltus raising $225 million for grid flexibility. Type One Energy raising $200 million for fusion. Axiom's $2 billion for power and cloud infrastructure. And SET and FEV saying robots that repair the grid are a priority. All of that is capital pooling around electricity. The AI boom needs enormous amounts of it, and the people building the machines to build and maintain the system are being funded alongside the people generating it. I'd be careful not to over read. The sources don't show that any specific robotics company is capturing a specific slice of the fusion or grid budget. But the direction of travel is clear, and it's consistent with Edison's Ziegler listing utilities, manufacturing, supply chain, agriculture, and transportation as targets where, in his words, return on investment is measurable through predictive maintenance, risk management, asset integrity, security, and autonomous operations.

Zach Martin

So we've seen the money, the splitting of brains and bodies, and the move toward real work. Now let's talk about what could go wrong, because a responsible episode about capital has to look at the graveyard. PitchBook's Ali Javaheri wrote in the Q1 report that the quarter also raised the diligence bar. His line is that investors no longer judge robotics companies on whether the technology works. They judge whether a company can manufacture at scale, win paying customers, and cut integration costs. That's a big shift. A few years ago, a good demo video could carry a funding round. Now the demo is the entry fee. The questions are about unit economics, payback periods, and how much it costs to put the machine into a customer's actual building.

Zach Martin

PitchBook also said that shift is moving capital toward industrial robotics, defense autonomy, and robotics software platforms, and that it raises harder questions for humanoid companies still leaning on demos, research buyers, and unproven paths to commercialization. Read that carefully. It's not saying humanoids are failing. It's saying the bar is higher for them, because their customers so far include research buyers, and a research buyer's willingness to pay isn't the same as a factory's willingness to rely on the machine every shift. PitchBook also pointed to the first meaningful exit wave, defense procurement marketplaces, and commercial robotics as a service contracts all landing in the same quarter. So there are signs of real deployment alongside the caution.

Zach Martin

Now the graveyard, drawn from the New Market Pitch table, and I'll be precise about what the data shows. Vicarious Surgical, which made miniaturized abdominal surgical robots, accumulated roughly $483 million across its private and direct rounds and is listed as shut down. Its last listed round was a $6 million registered direct and private placement in October 2025. Rethink Robotics, the collaborative robot pioneer, raised about $150 million and is listed as shut down, with its last round back in 2017. Rapid Robotics, which pitched robotic automation for manufacturing, raised about $54 million and is also listed as shut down. And on the other side, Auris Health raised $733 million and was acquired by Johnson and Johnson, which the brief says was a $3.4 billion upfront deal. That's the contrast the brief draws between liquidity and liquidation. Two companies can raise hundreds of millions, and one ends in a big sale and the other in a wind down.

Zach Martin

The lesson isn't that robotics is a bad bet. It's that equity dollars aren't a proxy for survival. Rangwala makes the same point in a different way. Physical AI takes patience, data, and extensive testing before it generates revenue. In software, you can ship a bug fix on a Tuesday. In hardware, if a gripper fails after 10,000 cycles in a customer's freezer, you may be shipping parts and technicians. The cost structure is different, the failure modes are different, and the investors who funded the 2021 generation of robotics learned that, expensively. Eclipse's Fath described the shift as funding moving away from experimentation and toward companies that can hit production milestones, land customers, and scale efficiently. That's the same lesson learned the hard way.

Zach Martin

That brings me to the three uncertainties from the briefing, which I think are the right questions to carry out of this episode. Uncertainty number one, which we touched on. How much of Skild AI's reported $100 million in annual recurring revenue represents sticky, multi year commercial software contracts, versus upfront enterprise pilot commitments? If it's the former, Skild is proving the horizontal brain thesis faster than anyone expected. If it's the latter, that revenue number could look different in a year. I don't know which it is, and the supplied sources don't tell us. What I'd watch is whether customers expand deployments over time, and whether the company discloses retention data.

Zach Martin

Uncertainty number two. Can humanoid pure plays like Apptronik and Figure achieve field mean time between failures that is acceptable to commercial warehouse and manufacturing clients before they consume their balance sheets? Let me translate that. Mean time between failures is how long a machine runs, on average, before something breaks. In a factory, if a robot needs rescuing every few hours, it's not saving labor, it's creating a new job. Figure has raised more than $1.7 billion, and the brief puts its valuation at $39 billion. Apptronik has raised $963 million. Those are war chests, but a war chest is also a clock. The question is whether reliability improves faster than the cash burns. The sources don't give failure data for either company, so this is purely an open question, and it's one customers will answer with their purchase orders.

Zach Martin

Uncertainty number three, the one I flagged earlier. Will pure play robot intelligence platforms, Skild, FieldAI, and Physical Intelligence, become commoditized horizontal layers captured by hyperscalers, or will vertically integrated hardware and software players establish deeper moats? Consider the evidence on both sides. For commoditization, we have the SET and FEV conclusion that base asset intelligence software has lost differentiation because AI lowered the barriers. If that happened to asset intelligence in energy, could it happen to robot brains? And we have chipmakers, AMD with World Labs and Nvidia with its Hugging Face bid, buying into models, which suggests the large platforms want the layer. For vertical moats, we have Eclipse's Fath saying the strongest moats go to those who own multiple layers, and Ziegler's idea that hardware is the distribution model for a data flywheel. I honestly can't tell you which side wins. Anyone who claims certainty here is selling something.

Zach Martin

Let me pull the threads together. Here's what the evidence supports. One, capital in 2026 is at record levels, with $679 billion in global venture through three quarters, and a record 27 billion dollar rounds in Q3 alone. Two, physical sectors, robotics, aerospace, semiconductors, data centers, and energy, each cleared ten billion dollars in the quarter. Three, this week's biggest deals sit underneath the robot: power, chips, design software, and manufacturing from code. Four, robot intelligence is being financed separately from robot bodies, in rounds that reach into the billions. Five, the industrial customers want machines that work across mixed fleets, and the energy investors want machines that fix things rather than merely watch them. Six, diligence has moved from demos to manufacturing, unit economics, and field reliability.

Zach Martin

And here's what's interpretation, which I want you to hold more loosely. That capital is rotating along a chain from software to power to silicon to machines. That chipmakers buying physical intelligence models is a strategic pattern. That OEM agnostic autonomy could win in heavy industry. And that the winners in energy robotics will be the ones who beat human crews on cost, as Austin Wood suggests. Those are reasoned reads of the sources, and I could be wrong about any of them. One more caution. Crunchbase's data are as of October 2, and early stage numbers get revised upward after a quarter ends, so the seed totals in particular will likely change. Treat this snapshot as a snapshot.

Zach Martin

If you want a single image to take with you, think about that Etched round. A chip company with $1 billion in booked orders, a first inference chip still in customer testing, no named buyers, no recognized revenue, and a $10.3 billion valuation that doubled in a matter of months. That's the entire physical AI market in miniature: enormous promise, enormous prices, and a gap between the order book and the proof. Some of these bets will turn into the next Waymo. Some will end up in the table of companies that shut down. The investors with the best odds, if the Q1 diligence reset is any guide, are the ones asking about manufacturing, payback, and failure rates, and not just whether the demo looks cool.

Zach Martin

That's Robot Capital for the week of October 12, 2026. Here's what I'll be watching in the coming weeks. Whether Flex follows through on separating Axiom as a public company early next year. Whether more companies disclose recognized revenue rather than booked orders or pilots. Whether the humanoid funding keeps flowing mostly to China, and what Unitree's public listing does for the sector. And whether the SET and FEV thesis about robots doing physical work on the grid starts showing up in actual funding rounds for field repair companies. Thanks for spending the hour with me on KMKT IR Hub Radio. I'm Zach Martin. Keep an eye on the money, and keep an eye on the machines. I'll talk to you next time.