The Physical Intelligence Economy Is Here. Most Investors Are Looking the Wrong Way.
The Physical Intelligence Economy Is Here. Most Investors Are Looking the Wrong Way.
Vitaly Golomb & Stefan Krause
Everyone is talking about AI. Almost nobody is talking about where AI actually matters most.
The past three years have been dominated by a single narrative: large language models, chatbots, digital AI agents, and the race to build the best foundation model. Hundreds of billions of dollars have flowed into this space. The valuations are staggering.
But here’s what the market is missing: the biggest AI opportunity isn’t digital. It’s physical.
Embodied AI—artificial intelligence integrated with robots, sensors, and machines that operate in the real world—is growing at 39% annually. Quarterly funding has tripled from $1.8 billion in Q1 2024 to $5.8 billion in Q3 2025. Figure AI just raised $1 billion at a $39 billion valuation with Microsoft, NVIDIA, and OpenAI all writing checks.
And yet, most investors are still staring at their screens, optimizing for the next chatbot wrapper.
We think they’re looking the wrong way.
What Is the Physical Intelligence Economy?
Let’s define our terms.
The Physical Intelligence Economy is what happens when AI leaves the data center and enters the real world. It’s robots that can see, feel, and make decisions in real time. It’s machines that learn from their environment without phoning home to the cloud. It’s intelligence embedded in physical systems—on factory floors, in operating rooms, inside warehouses, on roads.
This isn’t your father’s robotics. Traditional industrial robots were dumb. They repeated the same motion a million times. They couldn’t adapt. They couldn’t learn.
The new generation is fundamentally different. Foundation models like Physical Intelligence’s π0 can be trained across eight different robot platforms with just 1–20 hours of training data. The old approach required thousands of hours. These models combine visual perception, spatial reasoning, and motor control. They interpret natural language instructions and convert them into physical actions.
This is not incremental. This is a platform shift.
Follow the Money. It’s Moving Fast.
The smart money has already figured this out.
Quarterly robotics funding went from 89 deals worth $1.8B in Q1 2024 to 195 deals worth $5.8B in Q3 2025. That’s not a trend—it’s a stampede.
Look at the names on the cap tables: Microsoft, NVIDIA, OpenAI. These aren’t passive bets. Microsoft is integrating Azure with Figure AI’s humanoid robots. NVIDIA is supplying specialized hardware across the sector. Physical Intelligence raised over $400M in its Series B. FieldAI pulled in $405M for agricultural robotics.
But here’s the thing most people miss about these numbers: they’re concentrated at the top. The mega-rounds get the headlines. The real opportunity is one layer down—at the Seed and Series A stage, where specialized companies are solving specific physical problems with real unit economics.
That’s where the 10x–80x returns will come from. Not from the generalist platforms burning billions to build a humanoid that can fold laundry.
The Three-Layer Stack That Makes It All Work
To understand why embodied AI is suddenly viable, you need to understand the technology stack. It has three layers, and all three have crossed critical thresholds in the past 18 months.
Layer 1: Perception and Sensing. Multi-modal sensor systems now combine cameras, LiDAR, and tactile sensors in a single package. High-resolution cameras deliver real-time scene understanding. LiDAR produces 3D maps with millimeter accuracy. Tactile sensors detect pressure changes for precise force control. This is how a robot “feels” the world.
Layer 2: Intelligence and Processing. Foundation models provide semantic understanding and reasoning. Reinforcement learning enables continuous improvement. And critically, edge computing now achieves response times under one millisecond—so the intelligence runs locally on the robot, not in a data center. No latency. No cloud dependency.
Layer 3: Action and Control. Advanced actuators deliver force control accuracy within 0.1 Newtons and response times under 10 milliseconds. Modern manipulation systems have over 20 degrees of freedom with force-sensitive feedback. Research from Columbia Engineering and the University of Bristol shows 40% improvement in delicate object handling.
When you stack these three layers together, you get machines that can perceive their environment, reason about what to do, and act with precision—all autonomously, all in real time.
That’s the unlock. And it’s why the growth rate of this market—39% CAGR—outpaces early cloud computing (22%) and mobile internet (28%).
Where the Value Is: Four Markets That Are Already Moving
The data from actual deployments is no longer theoretical. It’s operational. Here’s what’s happening right now across four sectors.
Manufacturing ($520B TAM). This is the most mature application area, and the numbers are hard to argue with. Facilities report 30–180% throughput increases. AI-powered inspection systems cut defect rates by 60%. Automated systems reduce workplace accidents by 90%. Figure AI has already partnered with BMW for manufacturing deployment. Tesla is deploying robots internally. These aren’t pilot programs—they’re production.
Healthcare ($380B TAM). The surgical robotics market alone is projected to exceed $70B. Clinical studies show a 40% decrease in complications for certain robotic-assisted procedures. But the bigger story is elder care: Japan anticipates a shortage of 640,000 caregivers by 2040. Globally, the 65+ population will grow from 10% to 16% by 2050. These aren’t problems you can solve with software. You need physical machines in physical spaces.
Logistics ($290B TAM). Amazon’s robotic systems boost processing speed by 75%. DHL increases capacity by 40%. Error rates drop below 0.1%—compared to 1–3% for manual operations. Autonomous delivery robots now operate in over 100 cities. Cost reductions of 20–35% are standard.
And here’s the number that explains everything: the U.S. alone has over 600,000 unfilled manufacturing jobs. This isn’t automation replacing workers. It’s automation filling gaps that workers can’t fill.
The demand is structural. Demographics drive it. And it’s not going away.
The Geography Problem Nobody’s Talking About
Here’s a number that should wake up every U.S. investor: China installed 276,288 industrial robots in 2023. The United States installed 55,589.
That’s a 5-to-1 ratio.
This gap reflects something deeper than manufacturing capacity. It’s a strategic divergence. China is betting that physical intelligence is the next platform—and backing that bet with deployment at scale.
The U.S. has the research advantage. The best foundation models, the best AI talent, the best startups. But deployment lags dramatically. This creates both a risk and an opportunity.
The risk: if U.S. companies don’t scale production, the manufacturing and deployment advantage tilts permanently toward Asia.
The opportunity: the companies that bridge the gap between American AI research and scalable physical deployment will be enormously valuable. This is exactly where venture capital should be focused.
What’s Still Hard (And Why That’s Good for Investors)
We’re not going to pretend this is all figured out. There are real challenges.
Dexterity is still limited. Current systems achieve only a fraction of human manipulation capabilities. Many applications requiring fine motor control remain out of reach.
Reliability in unstructured environments requires extensive safety systems. Most deployments still involve some human oversight.
Energy efficiency and battery life remain constraints for mobile platforms.
System costs still restrict deployment to high-value applications. Broader adoption depends on manufacturing scale.
But here’s the investor insight: these are engineering problems, not fundamental barriers. They’re the kind of problems that get solved when talent and capital converge on a sector. And they create moats for the companies that solve them first.
Every challenge listed above is also a startup opportunity. The company that cracks reliable dexterous manipulation owns healthcare robotics. The company that solves energy-efficient mobile autonomy owns logistics. The company that achieves reliable unstructured-environment operation owns construction and agriculture.
The challenges are features, not bugs. They’re what separates the investable companies from the science fair projects.
The Reallocation Is Coming
Here’s what we think happens next.
The AI investment landscape is going to rebalance. The digital AI market is becoming crowded, commoditized, and increasingly competitive. Margins are compressing. Differentiation is shrinking. The best LLM wrappers in the world are struggling to build moats.
Meanwhile, embodied AI companies are building real defensibility: hardware-software integration, proprietary data loops from physical deployments, regulatory moats, and manufacturing know-how that takes years to replicate.
MarketsandMarkets projects the embodied AI market will hit $23 billion by 2030. Early deployments already show payback periods of 12–18 months in manufacturing and 24 months in healthcare. The unit economics work today, not in some hypothetical future.
At Mavka Ventures, this analysis drove our strategic focus on embodied AI. As the venture arm of Mavka Capital, a Silicon Valley investment bank specializing in frontier technology, we support companies building practical physical intelligence solutions—companies with real revenue, real customers, and real paths to scale.
We’ve spent 25+ combined years building startups, managing over $20B in M&A transactions, and investing in frontier tech. We’ve seen platform shifts before. Cloud. Mobile. Social.
This one is bigger. Because this time, AI isn’t just changing how we process information. It’s changing how the physical world works.
The physical intelligence economy is here. The question is whether you see it.
Everyone is talking about AI. Almost nobody is talking about where AI actually matters most.
The past three years have been dominated by a single narrative: large language models, chatbots, digital AI agents, and the race to build the best foundation model. Hundreds of billions of dollars have flowed into this space. The valuations are staggering.
But here’s what the market is missing: the biggest AI opportunity isn’t digital. It’s physical.
Embodied AI—artificial intelligence integrated with robots, sensors, and machines that operate in the real world—is growing at 39% annually. Quarterly funding has tripled from $1.8 billion in Q1 2024 to $5.8 billion in Q3 2025. Figure AI just raised $1 billion at a $39 billion valuation with Microsoft, NVIDIA, and OpenAI all writing checks.
And yet, most investors are still staring at their screens, optimizing for the next chatbot wrapper.
We think they’re looking the wrong way.
What Is the Physical Intelligence Economy?
Let’s define our terms.
The Physical Intelligence Economy is what happens when AI leaves the data center and enters the real world. It’s robots that can see, feel, and make decisions in real time. It’s machines that learn from their environment without phoning home to the cloud. It’s intelligence embedded in physical systems—on factory floors, in operating rooms, inside warehouses, on roads.
This isn’t your father’s robotics. Traditional industrial robots were dumb. They repeated the same motion a million times. They couldn’t adapt. They couldn’t learn.
The new generation is fundamentally different. Foundation models like Physical Intelligence’s π0 can be trained across eight different robot platforms with just 1–20 hours of training data. The old approach required thousands of hours. These models combine visual perception, spatial reasoning, and motor control. They interpret natural language instructions and convert them into physical actions.
This is not incremental. This is a platform shift.
Follow the Money. It’s Moving Fast.
The smart money has already figured this out.
Quarterly robotics funding went from 89 deals worth $1.8B in Q1 2024 to 195 deals worth $5.8B in Q3 2025. That’s not a trend—it’s a stampede.
Look at the names on the cap tables: Microsoft, NVIDIA, OpenAI. These aren’t passive bets. Microsoft is integrating Azure with Figure AI’s humanoid robots. NVIDIA is supplying specialized hardware across the sector. Physical Intelligence raised over $400M in its Series B. FieldAI pulled in $405M for agricultural robotics.
But here’s the thing most people miss about these numbers: they’re concentrated at the top. The mega-rounds get the headlines. The real opportunity is one layer down—at the Seed and Series A stage, where specialized companies are solving specific physical problems with real unit economics.
That’s where the 10x–80x returns will come from. Not from the generalist platforms burning billions to build a humanoid that can fold laundry.
The Three-Layer Stack That Makes It All Work
To understand why embodied AI is suddenly viable, you need to understand the technology stack. It has three layers, and all three have crossed critical thresholds in the past 18 months.
Layer 1: Perception and Sensing. Multi-modal sensor systems now combine cameras, LiDAR, and tactile sensors in a single package. High-resolution cameras deliver real-time scene understanding. LiDAR produces 3D maps with millimeter accuracy. Tactile sensors detect pressure changes for precise force control. This is how a robot “feels” the world.
Layer 2: Intelligence and Processing. Foundation models provide semantic understanding and reasoning. Reinforcement learning enables continuous improvement. And critically, edge computing now achieves response times under one millisecond—so the intelligence runs locally on the robot, not in a data center. No latency. No cloud dependency.
Layer 3: Action and Control. Advanced actuators deliver force control accuracy within 0.1 Newtons and response times under 10 milliseconds. Modern manipulation systems have over 20 degrees of freedom with force-sensitive feedback. Research from Columbia Engineering and the University of Bristol shows 40% improvement in delicate object handling.
When you stack these three layers together, you get machines that can perceive their environment, reason about what to do, and act with precision—all autonomously, all in real time.
That’s the unlock. And it’s why the growth rate of this market—39% CAGR—outpaces early cloud computing (22%) and mobile internet (28%).
Where the Value Is: Four Markets That Are Already Moving
The data from actual deployments is no longer theoretical. It’s operational. Here’s what’s happening right now across four sectors.
Manufacturing ($520B TAM). This is the most mature application area, and the numbers are hard to argue with. Facilities report 30–180% throughput increases. AI-powered inspection systems cut defect rates by 60%. Automated systems reduce workplace accidents by 90%. Figure AI has already partnered with BMW for manufacturing deployment. Tesla is deploying robots internally. These aren’t pilot programs—they’re production.
Healthcare ($380B TAM). The surgical robotics market alone is projected to exceed $70B. Clinical studies show a 40% decrease in complications for certain robotic-assisted procedures. But the bigger story is elder care: Japan anticipates a shortage of 640,000 caregivers by 2040. Globally, the 65+ population will grow from 10% to 16% by 2050. These aren’t problems you can solve with software. You need physical machines in physical spaces.
Logistics ($290B TAM). Amazon’s robotic systems boost processing speed by 75%. DHL increases capacity by 40%. Error rates drop below 0.1%—compared to 1–3% for manual operations. Autonomous delivery robots now operate in over 100 cities. Cost reductions of 20–35% are standard.
And here’s the number that explains everything: the U.S. alone has over 600,000 unfilled manufacturing jobs. This isn’t automation replacing workers. It’s automation filling gaps that workers can’t fill.
The demand is structural. Demographics drive it. And it’s not going away.
The Geography Problem Nobody’s Talking About
Here’s a number that should wake up every U.S. investor: China installed 276,288 industrial robots in 2023. The United States installed 55,589.
That’s a 5-to-1 ratio.
This gap reflects something deeper than manufacturing capacity. It’s a strategic divergence. China is betting that physical intelligence is the next platform—and backing that bet with deployment at scale.
The U.S. has the research advantage. The best foundation models, the best AI talent, the best startups. But deployment lags dramatically. This creates both a risk and an opportunity.
The risk: if U.S. companies don’t scale production, the manufacturing and deployment advantage tilts permanently toward Asia.
The opportunity: the companies that bridge the gap between American AI research and scalable physical deployment will be enormously valuable. This is exactly where venture capital should be focused.
What’s Still Hard (And Why That’s Good for Investors)
We’re not going to pretend this is all figured out. There are real challenges.
Dexterity is still limited. Current systems achieve only a fraction of human manipulation capabilities. Many applications requiring fine motor control remain out of reach.
Reliability in unstructured environments requires extensive safety systems. Most deployments still involve some human oversight.
Energy efficiency and battery life remain constraints for mobile platforms.
System costs still restrict deployment to high-value applications. Broader adoption depends on manufacturing scale.
But here’s the investor insight: these are engineering problems, not fundamental barriers. They’re the kind of problems that get solved when talent and capital converge on a sector. And they create moats for the companies that solve them first.
Every challenge listed above is also a startup opportunity. The company that cracks reliable dexterous manipulation owns healthcare robotics. The company that solves energy-efficient mobile autonomy owns logistics. The company that achieves reliable unstructured-environment operation owns construction and agriculture.
The challenges are features, not bugs. They’re what separates the investable companies from the science fair projects.
The Reallocation Is Coming
Here’s what we think happens next.
The AI investment landscape is going to rebalance. The digital AI market is becoming crowded, commoditized, and increasingly competitive. Margins are compressing. Differentiation is shrinking. The best LLM wrappers in the world are struggling to build moats.
Meanwhile, embodied AI companies are building real defensibility: hardware-software integration, proprietary data loops from physical deployments, regulatory moats, and manufacturing know-how that takes years to replicate.
MarketsandMarkets projects the embodied AI market will hit $23 billion by 2030. Early deployments already show payback periods of 12–18 months in manufacturing and 24 months in healthcare. The unit economics work today, not in some hypothetical future.
At Mavka Ventures, this analysis drove our strategic focus on embodied AI. As the venture arm of Mavka Capital, a Silicon Valley investment bank specializing in frontier technology, we support companies building practical physical intelligence solutions—companies with real revenue, real customers, and real paths to scale.
We’ve spent 25+ combined years building startups, managing over $20B in M&A transactions, and investing in frontier tech. We’ve seen platform shifts before. Cloud. Mobile. Social.
This one is bigger. Because this time, AI isn’t just changing how we process information. It’s changing how the physical world works.
The physical intelligence economy is here. The question is whether you see it.