Capital
South Park Commons has backed humanoid robots, autonomous satellites, and the data layer for robotic surgery, often before the founders had a product, a market, or even a company name. Dylan, investor at South Park Commons, talks with Ilir about why technical depth can’t be taught but storytelling can, why most PhD spinouts fail before they start, and why he’s not worried about warehouse robotics.
DYLAN is an investor at South Park Commons, a 10-year-old technical community turned early-stage fund with hubs in San Francisco, New York, and Bangalore, providing early-stage venture capital into hard tech and physical AI.
Give me the short version of what South Park Commons actually does.
We started as a community for the most curious technical minds to explore the frontier, a lot of early machine learning research, some blockchain. People from that early cohort went on to found Anthropic, OpenAI, the Ethereum Foundation. Post-GPT, our thesis shifted. Software is commoditized now. We’re more interested in the intersection of atoms and bits, not necessarily hardware, but things that are harder to build precisely because AI made software easy. We bring technical people into three physical spaces, then we’re the first check into teams building hard companies who are willing to sign up for the difficulty from day zero. We’ve funded humanoid robots, data center robots, drones, surgical robotics, and the software layer underneath all of it.
When there’s no product yet, sometimes no company yet, how do you actually evaluate someone?
We underwrite depth and clarity, the ability to train foundation models, a real pedigree in mechanical engineering. But having them in our community for a while matters more. LOIs aren’t that valuable to me. What’s interesting is what they can build with less. We have a team going for FDA approval on a surgical data-capture device, most investors won’t touch anything that needs FDA approval. We have another team building embedded cameras for construction hard hats, but instead of building custom hardware for years, they proved value with off-the-shelf gear first. And we backed a humanoid robotics team that built two fully autonomous robots, not deployed yet, but working, live, not a demo video nobody can verify. That’s the difference: proving you can do one or two of these before you ask for the money to scale. There are other ways to earn that credibility, too. The founder of our satellite company was one of five people on Earth who had done autonomous docking in space, from his time at Northrop Grumman. He didn’t need a demo, he’d already proven it in another capacity.
What actually separates a founder with a great idea from an exceptional founder?
Our bet is that you can’t teach execution, but you can potentially teach storytelling. If we find technically excellent builders who can actually execute, can we help them get better at world-building? The founders of Physical Intelligence were very academic when they started, they learned how to be CEOs, how to fundraise. We’re looking for people who spike on technical depth and execution, hoping we can bring the storytelling up to a level where it’s not embarrassing. It doesn’t have to be a pitch, a white paper, a demo, a hardware prototype all count. It’s proving you can build and articulate something that doesn’t exist yet. We use a framework internally: a tell me founder versus a show me founder. If storytelling isn’t your superpower, become a show me founder instead, get as creative as you can with what you can actually demonstrate, even without the money for a real product yet. Don’t let perfection kill momentum.
What’s the single biggest thing that keeps brilliant researchers from ever leaving academia to found something?
A tenured professor finds a postdoc, convinces them to spin the lab’s research into a company, takes a chunk of equity as co-founder, and lets the postdoc do all the work and be the face of it. It’s a free lottery ticket for the professor, cheap to test, low cost if it fails. That pushes postdocs into founding before they’re ready. Some of them are ready. Most would be better off getting real operating experience first, hiring, managing, shipping, and then going back to their old advisor later if it still makes sense. We funded a CMU PhD who went to work at NVIDIA for a couple of years first, then came back to his advisor to build the company. He’s better for it.
“Venture capital is risk capital. The question is: how many miracles need to happen for this to work?” South Park Commons’ framework for underwriting hard tech; count the miracles required, then find the team that only needs one.
How do you know a team can cross from an impressive demo to real deployment?
We ask how many miracles are required. Can they de-risk the technical miracle, can they build what they say they’ll build? That’s fine if it’s the only one. It gets harder when there’s also no market yet. Software for robotics companies that don’t exist yet is two miracles. We want teams where the only open question is technical, where if I’m convinced the demand is real, the only bottleneck is that nobody has built this affordably before, and I believe they can.
Robotics is capital-intensive and slow. What has to be true for you to make that trade-off versus a pure software bet?
Not every hard tech company needs to build its own hardware. A lot of it can be dumb, off-the-shelf hardware, the value is in the operating system, the data, or the simulation environment on top. Our satellite company has to build its own hardware because nobody else builds reusable autonomous-docking satellites. Our surgical data company doesn’t, they don’t want to perform surgery, they want better data for autonomy that’s already happening. Founders need extreme clarity on why they need to build their own hardware, because that’s a much bigger capital ask. We’re happy to underwrite technical depth with a first check of one to five million dollars, but that check alone won’t get a hard tech team very far. Our satellite company, if it works, will need to raise something like four to five hundred million dollars over the life of the company. So it’s not just “can they build it”, it’s “can they also raise at that scale,” and that’s something we can coach on but they ultimately have to be great at themselves. I’m less excited about generic warehouse and packaging robotics, that’ll probably end up being one or two large players, maybe Amazon and Tesla. I’m more interested in high-cost, high-skill, low-supply labor: welding, precision manufacturing, composite work. And I’m not a believer in consumer robotics yet, the willingness to pay isn’t there.
You call yourselves a community, not just a fund. Is that actually an investment edge, or is capital still the differentiator?
Capital is a commodity, everyone’s investing in hardware now. The community is the edge. We’ve seen people come in wanting to found something, realize they’re not meant to be founders, and end up as head of hardware or CTO at one of our other companies instead. Because everyone’s been through our space, they become customers of each other, they share insight with each other. And when we’re evaluating a team outside our own expertise, we can ask other PhDs in the community what they think of the approach. That’s a depth of diligence capital alone doesn’t buy.
You could move up-market into Series B and C, where the checks are bigger. Why stay this early?
We take all the risk, we’re investing when there’s the least to go on. That builds a different kind of relationship with founders than their later investors get. It’s not that later money feels transactional exactly, but there’s more at stake for those investors, and founders sense it. Our founders come back to us first when things aren’t going well, because they feel safe doing that. The other thing I’ve learned moving from software into hardware is that the feedback cycle is so much longer. With software you get customers and revenue fast and you can raise again fast. Hardware takes way longer, it could be a year, could be a year plus before a team is in a position to raise more. We have to be patient capital and let founders know we’re comfortable with that timeline, because I don’t think every investor who now says they want to do hard tech actually knows how long this stuff takes.
Where should ambitious researchers be spending the next six to twelve months?
Getting out of the lab and talking to customers, not necessarily chasing LOIs, but validating that what they’re building is actually needed and that a hundred other people aren’t doing the same thing. That used to matter less; the field is far more crowded now. I’d also pay attention to what the major labs might do in robotics themselves, Anthropic, OpenAI, Amazon, Tesla have all signaled intent here, so know what’s uniquely yours. And drop the obsession with building your own hardware when you don’t need to. Some of our best teams have this instinct to build everything themselves because it’s an easier problem than walking into a factory and talking to people they’ve never met. Don’t reinvent the wheel, repurpose it.
Update:
Since this conversation, South Park Commons has closed its largest fund to date: $575 million for Fund IV, more than double the $275 million Fund III, bringing assets under management to roughly $2 billion.