Saman Farid grew up in Beijing. His parents moved there when he was six, and he watched China go from dirt roads to the manufacturing capital of the world. He built his first company at 14. He later built and sold a company in China, then spent a decade as a robotics investor at Baidu, where he put more than $650 million into over 40 robotics companies. Along the way he decided the biggest problem in robotics was not the robots.

Today he runs Formic. It operates the largest independent robot fleet in the United States, with over 500,000 production hours across more than 100 factories. Formic owns and operates the robots itself and charges customers for performance, not hardware. Its bet is that the hard part of robotics was never making a robot work in a lab. The hard part is getting it deployed, reliably, at scale, inside a real factory.

This matters if you build or buy Physical AI. Most of the money in the field is chasing humanoids and foundation models. Farid has spent years in the dirty, dusty, dangerous parts of manufacturing where robots actually earn their keep, and his case is blunt: deployment and distribution beat raw intelligence. The future of automation runs through the factory floor, not the lab.

The conversation

Jay: You grew up in Beijing. You watched China go from dirt roads and donkeys carrying bricks to the manufacturing superpower of the world. Then you invested $650 million across 40 robotics companies at Baidu. And after all of that, you decided the biggest problem in robotics wasn't actually the robots. What did you see, and when did you realize the real problem was somewhere else?

Saman: My parents moved to China when I was six. We were the only non-Chinese family in my school and in the whole area we lived in. People would stop us on the street, surprised to see a foreigner. I feel extremely thankful that I got to experience China going through this massive industrialization, where incrementally you could see every aspect of people's lives improve as the industrial base got stronger. People's daily lives depend on their society's ability to produce the things they use and need.

Later, as an entrepreneur, I built factories in China and worked with a lot of manufacturing and technology companies there, and I saw the rate at which adopting new technology pushed society forward. There's a cliche in Silicon Valley: the future is here, it's just not evenly distributed. That's extremely true in robotics. Go to any robotics lab, or just go on YouTube, and you can see robots doing backflips and all kinds of tricks. Go to any factory, construction site, or farm, and there are no robots to be seen. That gap was the big thing in my mind. As an investor I was funding new robot capabilities every day, but there was a massive chasm when it came to adoption in the real world. What excites me in technology is diffusion. The incremental new trick is nice. Getting it into people's lives, where it actually helps them live better and make better things, is what matters.

Jay: Let's put numbers on that chasm. Where is the capital actually going?

Saman: We did a calculation recently. In the last 12 months, about $27 billion went into physical AI and robotics. About 95% of that went into companies doing either humanoid work or foundation model work. Of all the companies funded, only a handful have actually deployed robots into the real world, and a very small handful have deployed to more than 20 or 30 sites.

Jay: So robotics is stuck in pilot purgatory, where the capital going in doesn't line up with the revenue opportunity. Why? What is the industry getting wrong, and how do you see it differently at Formic?

Saman: People are trying to pattern match what happened with LLMs onto robotics. It's seductive, but it's a mistake. With LLMs, the models weren't very good until ChatGPT came out and people realized it was actually working. Then you had this massive takeoff in adoption, and there was real revenue behind those businesses.

In robotics, people ask, when is the ChatGPT moment? I think that fundamentally misunderstands the industry. The narrative is that once the robot works, adoption happens instantly. My observation over the last 10 years is the opposite. Getting the robot to work is the easy part. Go to any lab or any pitch event in the Bay Area and you'll see startups with robots that pick strawberries, fold laundry, till soil, pull weeds. Any use case you can imagine, someone has built a robot that does it pretty well. Adoption is the hard part.

I can say this as Formic, because we've deployed hundreds of robots in hundreds of factories, and we own and operate all of them. The time and energy it takes us to program the robots is less than 5% of our deployment cost. So say magically tomorrow the robotics foundation models were perfect and worked every time. You've reduced my programming time from 5% to maybe 1%, because I still have to run them. That's inconsequential for a business deploying lots of robots. I'm enthusiastic that these models are getting better, but there won't be a single ChatGPT moment on the model side.

Jay: You've said the robot is only 10% of the problem. The other 90% is error handling, maintenance, programming, change management. Give me an example that would make a software investor's head spin. What's the most absurd failure case you've seen when the plan meets first contact with the enemy?

Saman: I have so many. I'll give two, because they're different. One is fundamental physics. A robot is designed with a certain capability set. This arm, these motors and actuators and grippers, can pick up something up to about 30 pounds and move at a certain maximum speed. Those constraints don't go away if the intelligence gets better. So a factory might run one SKU all day, and then on Sundays run a different SKU that's heavier or bigger or faster, and suddenly the robot can't keep up.

The other one, since you asked for absurd: early on, we'd scope a solution, get everyone aligned, and the factory would buy into the plan. Then on deployment day we'd show up with the truck, roll the robot off, wheel it toward the room where it's supposed to work, and the door in between was too narrow to fit the robot through. We added it to the checklist. But there are a hundred little things like that. Is there a column in the way? Direct sunlight that washes out the camera? Is the bag you're picking porous or hard? There are maybe 500 pieces of information you need before you can design, scope, and deploy a robot successfully. The robot being smarter doesn't absolve you of that work.

Jay: Where does that show up in the cost structure? You grew deployments 5x year on year and brought down cost per deployment. What has this past year taught you about building a high-margin hardware business?

Saman: A couple of foundational principles, none of it rocket science. One, remove customization as much as possible, get it close to zero. Two, build modular, standardized systems that don't require customization for different tasks. Think LEGO blocks you mix and match and deploy. In traditional robotics, about 60% of deployment cost is manual hardware configuration and integration. How do I connect to this conveyor? To their ERP? What about safety infrastructure? A hundred questions, and doing a custom design for every task isn't feasible when you're scaling fast. So at Formic we compressed all of that into software. We built a tool called Formic Cortex that analyzes the requirements of the task, selects the standard modules we need, validates them in simulation, and then uses cameras in lieu of basically all other sensors. Instead of integrating with the conveyor or the machinery upstream and downstream, the robot just observes what's coming, like a human does.

Jay: So no PLC integration with their existing conveyor. It's all vision-based. You show up ready to go out of the box.

Saman: That's right.

Jay: Let's talk business models. A couple of years ago the thing that got investors excited about robotics was robots as a service. It looked like SaaS. You're doing something more unique, utilization-based and success-based. Why is that a better model?

Saman: I'll push back a little. It doesn't matter what's better for me as a venture-backed company. It matters what's better for my customer. My firm belief is that having skin in the game on your customer's production output, having aligned incentives, ripples into every decision you make. The fundamental question is, if I were the customer, what would I want? Would I want someone to sell me a few hundred thousand dollars of capex and then have no responsibility after? Or someone whose feet I can hold to the fire on performance? Obviously the second.

The second layer is: if robots as a service is better for the customer, how do you actually do it as a business and survive? There are examples to learn from. Caterpillar has power by the hour, where you pay for the performance of your equipment. Rolls-Royce does it with engines. AWS lets you pay for utilization. A lot of it is having very good predictive models for the future utilization of an asset. Over the last few years we've built a model that predicts the expected utilization of an asset in a certain environment, with a certain credit-quality customer, a certain use case. It turns out robots are a very good asset class. We're generating roughly 65% net asset IRRs on these robots. Late last year we did our first securitization. We pooled these assets together and had equity and debt investors buy in.

Jay: A lot of people misunderstand robots as a service as just a different pricing model.

Saman: Either I price it at 100,000 up front or 30,000 a year. People who go that route tend to really regret it. I've seen many startups offer robots as a service for a few years, run out of cash, and go back to capex sales, because if you don't build the infrastructure to fund it, you can't scale.

Jay: Are you still seeing a disconnect between what robotics companies want to sell and what customers want to buy?

Saman: One layer of nuance: it's not that they don't have capital. They buy forklifts all the time. They build new buildings all the time. What they don't have capex for is something they perceive as risky, uncertain, or outside their expertise. Robotics falls squarely into that. I run a chocolate chip cookie factory. I know everything about chocolate chip cookies. I'm not likely to also be an expert in robotics. So when you show me this newfangled robot, my first reaction is, that's cool, I want it, but how do I know it'll work?

One of our sales pitches is that we help you avoid the tarp of shame. The tarp of shame is a term we invented. It's very common for a factory to spend a bunch of money on a piece of equipment they think is awesome, it works for a week, and then it sucks. It slowly gets pushed into the corner of the factory, they cover it with a tarp, and they never talk about it again. Talk to any factory owner about the tarp of shame and they'll immediately laugh and say they have three of them in the back. So really this is about risk. Who should own which part of the risk? Robots as a service is a better proposition because it says: don't worry about how much it costs, don't worry about keeping it running, only pay for performance.

Jay: That's a great deal.

Saman: The main directive I've given my sales and product teams is: invent a product that you'd have to be stupid to say no to. Sometimes our salespeople go to a customer's CFO and say, you'd be financially irresponsible not to buy this. Just look at the numbers. That's the goal of every startup.

Jay: You do some aggressive things. A three-month trial, a no-fault money-back guarantee. Has anyone sent a robot back, and what did that teach you?

Saman: We've had customers send them back, usually because utilization wasn't high enough to justify it. We do a lot of work up front to vet that an opportunity is worthwhile, and we drop out of deals all the time, often to the anger of our sales team. We won't over-promise and under-deliver. The result is a 97% renewal rate, so about as close to zero churn as you can get in a startup, and a lot of expansion. Our net dollar retention is around 180%. Every customer starts with a couple of robots, sees it works, and expands.

Jay: You have three customer profiles. The third-generation family business. The PE-owned portfolio company. And the first-time manufacturer, the American re-industrialization story. What have you learned about selling into these three?

Saman: All of them care about one thing: how can I be more competitive and win more business? For legacy manufacturers in America, the vast majority are actually turning down business every day. They have opportunities to do more but can't meet the price point or the throughput.

One piece of context that helps: the labor gap is enormous. There are about 1.2 million unfilled manufacturing jobs in America right now. I visited a customer in New Jersey a few weeks ago. The factory has about 160 employees. In 2025 they had to hire 275 people just to keep it running. So there's no narrative here about robots taking jobs. The typical American factory runs about one third of its available production hours in a year.

Family-owned businesses are generally focused on not screwing things up. It's often third, fourth, or fifth-generation businesses, focused on maintaining what exists. These businesses are the backbone of their community, and they're very concerned that if they overextend, they go out of business. So risk mitigation is critical. Private equity is different. Shorter time horizon. It's about reducing opex quickly and increasing revenue quickly without taking on new capex. The third category, new plants spinning up, is about using robotics to win business quickly against incumbents.

Jay: When you stack rank them, who makes up the majority?

Saman: The family-owned businesses are probably the biggest segment. One stat people don't know: 98% of American manufacturing businesses are SMBs. A single GM or Tesla factory has about 5,000 suppliers building components they ship in to assemble a car. That 5,000-to-1 ratio is common in automotive. In aerospace it's more like 20,000 or 25,000 to one. For every big company, you have a few hundred or a few thousand small companies making the things that show up there.

Jay: Turnover in factories runs 200% a year. We've seen it higher, 500% a year in a recycling plant we're invested in. These are dirty, dusty, dangerous jobs. When you walk into a facility, how are the people still there thinking about their robotic coworkers?

Saman: People are very excited when robots show up. We throw a robot naming party, bring a birthday cake for the robot, the staff name it, and we stick the name on it. Generally these robots do the most painful, backbreaking, laborious tasks in the factory. For those really backbreaking jobs, there usually isn't one person doing it all day. You can't pick up 50-pound boxes for 12 hours. So they rotate. Everyone wants to skip that miserable part if they can. So when the robot comes in, everyone celebrates. The only people we've really heard the robots-taking-jobs question from are mostly reporters.

Jay: You wrote in Forbes that the humanoid hype is purely hype. Make the case.

Saman: The fundamental issue with humanoids is that it lacks creativity. If you were redesigning something from scratch, why keep it to two arms and not six? Why two legs that walk two miles an hour instead of wheels that go 60? You're making a lot of trade-offs just to make it look human. To steal a line from Steve Jobs, a human is one of the slowest animals on Earth, but a human on a bicycle is one of the fastest. We have the ability to drastically improve the form factor, so I don't understand why we'd choose not to.

Jay: Let me push back. Our homes have stairs, our factories have narrow doors. A six-armed wheeled spider-bot won't fit through the door, but a slender humanoid will. Doesn't that take feet?

Saman: Not necessarily, but it's a good point. There are tasks that only the human form factor can fit into or do, granted. So humanoid demand isn't zero. My main contention is you lose a lot in that choice. In a factory, a farm, or a construction site, you're choosing between superhuman capability in a slightly different form factor, or human-to-subhuman capability in a human-like form factor. We've made this trade-off many times in history. Look at a lawnmower. You don't build it in a human form factor and send it out with scissors to cut each blade of grass. In most B2B use cases, a robot does one or a few things all day, every day. You pick one thing and make it really good at it. When you do need some generalizability, you'll have gap-filler robots, and that's where a humanoid makes sense.

Jay: Let's talk national security. Say I make you the robotics czar for the US. What has to happen over the next 24 months for the US to take the top spot?

Saman: I advocate very strongly that good users of robots become good makers of robots. The first thing we need to do in America is use a lot more robots. That will in turn train the supply chain. In America, 90% of factories don't use any robots at all. So building America's robotics capability comes from being a very good user of robots.

The secondary thing: there's a lot of enthusiasm for investing in final robot manufacturing. But look at China's industrial policy. It took a layer-by-layer approach. In the 70s and 80s they incentivized basic materials, how to make steel, forges, nuts and bolts. Then they moved up to toys, simple assemblies, basic machinery, then smartphones, and only from there to advanced shipbuilding, drones, robots, solar panels, and semiconductors. These things are a pyramid. You start at the bottom. The US used to have this pyramid, but a lot of the bottom pieces have gone away. Now everyone's investing at the top. How are you going to make a robot if you can't make nuts and bolts? Where are the motors? Where are the magnets?

Jay: Let's close here. Five years out, what does the future look like?

Saman: We want to live in a world of abundance in every aspect of our material lives. There are billions of people on Earth who don't have enough of the basic necessities, and a lot of that comes down to supply chain efficiency. It's the boring answer. If we do our jobs well as a robotics industry, we can train a generation of manufacturers 10 or 100 times more productive than they are today. The alternative is that if the US doesn't do it, China will. It should be a wake-up call for American manufacturing that we've been resting on our laurels for the last 20 or 30 years. It's important for America to remember where we get all our stuff from. Look around this room. Every single thing here was made in a factory. My favorite activity lately is taking my son to Costco and pointing out all the stuff our robots help make. And it's worth noting, America is still the second-biggest manufacturer in the world. There are 250,000 factories here. There's just a lot of work to do to modernize it.

Pull quotes

1. "Getting the robot to work is the easy part. Adoption has actually been the hard part."

2. "The time and energy it takes us to program the robots is less than 5% of our deployment cost."

3. "Invent a product that you would have to be stupid to say no to."

4. "How are you going to make a robot if you can't make nuts and bolts?"

5. "Good users of robots become good makers of robots."

Source

From CLIMB Episode 094 with Saman Farid (Formic). Watch the full episode: https://youtu.be/mWdYfFUHAp4