Why the Best Physical AI Startups May Look Boring

Most of the capital going into physical AI today is chasing humanoid robots.

The investor I just spent an hour with will not fund a single one. And after listening to his reasoning, I think he is right.

I hosted Doug Griffin, Founder and Managing Partner of Spatial Capital, for our latest session. Doug spent about 30 years in computer vision before he started writing cheques. He worked in the R&D group at Industrial Light and Magic on Star Wars and Pirates of the Caribbean. He ran a technology team at Electronic Arts. He co-founded an animated film studio with Robert Zemeckis. He founded four companies and exited three, two of them to Apple and Roblox.

So when he says the industry is looking in the wrong direction, it is worth stopping to listen.

Here is what I took away.

The cameras did not get better. The software did.

I had this backwards before the conversation, and I suspect most people do.

Cameras have more resolution now. LiDAR is cleaner. There are more sensors everywhere. But Doug's point is that the sensing side has barely moved in a decade.

What changed is the AI stack sitting behind it.

Five years ago, if you wanted a robot arm to weld a car door, you programmed every single angle by hand. Then you controlled the room around it. The car had to arrive in exactly the same position every time, because the robot could not see.

Today the arm looks at the car and works it out.

Not because it has better eyes. Because it finally understands what it is looking at.

Why he will not fund humanoid robots

This was the most contrarian thing he said, and he knows it.

He gave two reasons.

The first is about jobs. Every startup disrupts something. Spatial Capital looks for the dirty and dangerous work nobody wants in the first place. A machine built in human shape is built that way for one reason, and he is direct about it.

"The reason they're building it like a humanoid is to replace human workers. And that's just not something we really want to do."

The second is about adoption, and I think this is the one founders underestimate.

"The thought of having a humanoid robot doing the dishes sounds really cool. But how many people would actually want a robot walking around their house, hearing everything they say, learning from it?"

Then he made a point that is almost too simple. If your robot works on a flat factory line, why does it need legs? Put it on wheels. Cheaper, simpler, better at the job.

He calls this fit for purpose.

What he funds instead: forests, bridges, mines and power grids

Spatial Capital's filters are narrow on purpose:

• Stage: pre-seed and seed, teams of three to twenty people
• Geography: North America and Western Europe
• Market: B2B only, no consumer
• Sectors: warehousing, agriculture, mining, energy, infrastructure inspection

His own description of the target: "old, ugly, boring businesses, things that run the economy but aren't that sexy."

One portfolio company, Deep Forestry out of Sweden, flies autonomous drones through forests and identifies every tree. Species, width, alive or dead, suitable for 2x4s or plywood. The data geolocates, so you can drop a whole forest onto Google Earth.

That serves two huge markets. Wildfire management, where measuring the fuel load in a forest is impossible on foot. And lumber, which touches every house built.

Nobody had that job before. Nothing is being replaced.

Another company, Cone Labs in Canada, reconstructs bridges in 3D from drone footage and finds cracks in infrastructure that is seventy years old and now meeting climate change.

None of this goes on a billboard. All of it has to happen every year, forever.

The real argument: software moats are dying

This is the part that should worry anyone building SaaS.

Doug is a former engineer who has not written a line of code in years. He has still built his own deal sourcing and grading tools by vibe coding them.

"You're going to see a lot of SAS businesses go away, because that moat, that barrier to entry, and the value they can charge, it's just not balanced out anymore."

Think about it from your customer's side. If the thing you sell for a monthly fee can be rebuilt in an afternoon for the price of a coffee, your pricing is on borrowed time.

Physical AI is different, and the reason is not glamorous.

The hard part was never the model. The hard part is going out into the world and capturing something nobody else has captured. Deep Forestry's moat is not clever code. It is years of flying drones through real forests before the model is worth anything.

His pattern for a fundable company:

1. Hardware that collects information about the world which is hard to capture
2. An AI model built around that data
3. Autonomy or action on top, serving a real business need

Each layer makes the next harder to copy. You cannot prompt your way around it.

He wants founders who already know the industry

At Series A you read the numbers. At pre-seed there are no numbers, so you read people.

Doug uses an old frame: the jockey, the horse and the course. The founder, the product, the market. Early on, the founder carries most of the weight.

He wants a technical founder, usually a PhD or deep AI background, paired with someone who actually came out of the industry they are selling into. Two technologists with no commercial read on the market is a pass, however good the demo.

"If you're doing a B2C startup, you can be straight out of college. But in B2B, you have to understand the businesses you're serving. You need to know their exact problems. You need to know how to sell into that industry."

I see this constantly in our programs. Selling into a mining company or a utility is not a pricing page. It is phone calls, procurement cycles and certifications. That knowledge cannot be learned quickly, and Doug wants it on the cap table rather than on the roadmap.

But a team that looks perfect on paper carries its own risk.

"If everyone comes from a Fortune 500 company, if they've never started a startup before in their life, it is a hard business starting something up. No one takes your phone calls. No one's giving you any respect."

Building is easy now. Commercializing is not.

This is the gap I see every week with deeply technical founders in Japan, Korea and Taiwan. They can build anything. They freeze when it is time to sell it, or even to post about it.

Doug put it plainly.

"We don't live in a world where if you build it they will come. Anyone can build a product very easily. It's understanding, does this product serve the market need? How can you reach those customers, and how can you convince them that you're the trustworthy partner they want?"

Four to eight words

The most useful exercise of the session took thirty seconds, and I will be stealing it for every program I run.

Describe your company in four to eight words. No company name. No "we are building". Just the problem and the product.

His example: "Agentic AI for property management." Five words.

That line goes under your company name on the title slide. Then slide one is the problem. Slide two is the solution. Slide three is the market size. Then the team.

"You don't want an investor eight slides in, still trying to figure out what you're doing."

When we ran this live with the audience, the entries that worked were plain and specific. "Automated manufacturing for custom textiles." The ones that failed hid behind big words. Vague is not sophisticated. It is just vague.

How to prove something before you have revenue

I asked him what counts as traction when you have no revenue yet. He ranked the proof points from weakest to strongest.

1. Respected angels from your industry. The president of an energy company putting 50K into your energy startup is a signal, not just a cheque.
2. A testimonial from an industry insider helping you shape the problem.
3. An unpaid proof of concept with a real customer. Meet every two weeks, ship a build, take the feedback.
4. A paid proof of concept. Charge 10K for a six month trial.
5. Service revenue while you build. Solve the problem manually for a customer while you build the product.
6. Recurring revenue. Even 10K to 15K a month makes a strong pre-seed company.

At seed he wants 10 to 15 percent month-on-month growth. The number can stay small.

His favourite signal costs nothing. A large company taking your call.

"If you can get a Fortune 500 company to say, yeah, we'll work with you if you build this, that tells me you've got business people smart enough to reach out to a big company and have them take the call. That's validation."

Do not send him a cold message

I asked whether he takes cold outreach. Almost never, and it actively puts him off.

"I've been around my industry for 25, 30 years. If you're in this, you probably know someone who knows me. Don't reach out cold. Find out who knows me, ask them for a warm introduction."

He was sharpest about AI-generated messages that scrape his LinkedIn and website and repeat it back. Spatial Capital keeps its website deliberately vague about the thesis, so the mirroring is obvious.

What works instead is slower. Warm introductions. Demo days. Accelerators. Conferences. Hosting your own events.

Exposure you build on purpose, not noise you spray at people.

The one question most founders cannot answer

I asked for a single lesson for founders. He answered with a question.

"Why are you in particular building that company? I need to know why this particular startup is going to beat another startup. There's inevitably going to be at least a dozen other startups in whatever sector you're looking at."

If you cannot answer that, you have two options. Recruit the people who can, as co-founders or advisors. Or go and work inside somebody else's startup until you become that person.

Neither of those is failure. Pretending you already have the answer is.

Watch the full session

This is the short version. The full conversation runs just over an hour and ends with a live Q&A where founders pitched their companies in four to eight words and got feedback on the spot.

Watch the full episode on YouTube "The “Old, Ugly & Boring” Physical AI Opportunity | Doug Griffin, Spatial Capital"

https://www.youtube.com/watch?v=TzCGZS8nLpg

Subscribe for more conversations with the founders, investors and operators building globally.

About Spatial Capital: Pre-seed and seed investment in physical AI across North America and Western Europe. B2B only. Portfolio includes Deep Forestry, Cone Labs and Vizonare.

About Openfor.co: An agentic entrepreneurship incubator supporting founders, operators and professionals with Silicon Valley grade access, programming and investor relationships.