Seth Berman

Seth Berman walks into Y Combinator demo days in 2010, and the room has twelve founders in it. Twelve investors. He ends up writing checks to about thirty companies over the next few years, including Nest, Optimizely, and a few others that won't become famous until later. He is not an engineer. He is not a banker. He spent the better part of a decade at Richemont Group - the Swiss luxury house that owns Cartier, Montblanc, Net-A-Porter, and Chloé - thinking hard about why people covet things, how brands create desire, and what makes a consumer relationship compound over time. That background turns out to be unusually useful for picking seed-stage companies.

In 2013, Berman co-founded Susa Ventures with Chad Byers and Leo Polovets. The name is a nod to the ancient city of Susa, where the first recorded financial transactions happened. The fund started small. The bet was simple: find companies with compounding moats - data advantages, network effects, proprietary datasets - and get in early enough to matter. Eighty percent of the portfolio went into four categories: logistics and sharing technology, fintech, digital health, and computer vision. It was a thesis that rewarded patience and specificity over dilettante generalism.

The early portfolio was studded with exactly the kind of outcome that thesis predicts. Expanse, a network security startup that scanned the entire internet to map corporate attack surfaces, was acquired by Palo Alto Networks for $800 million. Flexport became the defining logistics-tech company of its era. And then there was Robinhood.

The Robinhood story deserves its own paragraph. When the zero-commission brokerage came seeking seed capital, no institutional investor wanted to touch the regulatory complexity. To launch, Robinhood needed someone to be the investor of record for FINRA - to open their books and stake their credibility on a company promising to let everyday Americans trade stocks for free. Nobody volunteered. Berman did. He credits their eventual success to something that any good brand strategist would notice: "intuitive UI and great user experience." He kept 80% of his Susa stock through the IPO. That sentence tends to end conversations.

The compounding moat thesis wasn't just a phrase on a pitch deck. Susa built a scout network - people embedded at Robinhood, Stripe, Facebook, Google - tasked with identifying departing engineers and product managers before they hit the open market. While other funds waited for deck submissions, Susa was already in the room. The firm also built what it called a "parallel fund structure" - returning profits to portfolio founders rather than just to limited partners. It was a way of saying: we want to be the investors you call when you're starting your second company, too.

Berman's investment thesis has always centered on one question: what makes this company harder to displace next year than it is today? The answer, for most of Susa's portfolio, lives in data. Companies like Viz.ai ingested millions of CT scans to improve their stroke-detection AI. Flexport got better at logistics by moving more freight. The data moat grows with every transaction, every shipment, every scan. Berman called it early - "data was the new oil" - before that observation became a bumper sticker.

His newer framing is more specific, and potentially more important. At Kivu Ventures, his team coined "physical intelligence" to describe what they see as AI 2.0 - the moment when trained models stop living only in software and start affecting atoms. Autonomous vehicles, surgical robots, AI-designed chips, smart logistics nodes. The intelligence becomes physical. Berman is betting that this transition generates the next cohort of category-defining companies.

The seed market got crowded. By the time Berman talks about it publicly, there are 900-plus seed funds. He watched the stage compression in real time: what used to be seed is now Series A; what used to be pre-seed is now seed. The opportunity at Series A shrank to roughly 30 high-quality funds competing for the best deals. Berman moved up the stack.

Kivu Ventures targets companies that have demonstrated product-market fit - those that have cleared the existential questions and now need capital to scale a defensible advantage. Checks go up to $15 million per company. The focus: vertical AI, AI infrastructure, healthcare, fintech, logistics, and consumer tech. The logic follows the physical intelligence thesis directly.

The venture landscape Berman entered in 2010 bears little resemblance to the one that exists today. At early Y Combinator demo days, the room was small enough that everyone knew everyone. There was no crowd. Berman invested in roughly 30 companies in that era - several became unicorns - partly because there was genuine information asymmetry, and partly because being a consistent, reliable early participant built a reputation that compounded into dealflow.

That asymmetry is largely gone. Nine hundred-plus seed funds compete for the same deals. Winner-take-all dynamics mean that if a company is talking to 20 funds, 19 of them lose. Berman's prescription: be the partner founders want, not just the fund with the best platform deck. "When founders receive multiple tier-one term sheets, they tend not to care so much about the platform and the services. Instead, they prioritize the partner they're going to be working with."

The AI wave adds another layer of complexity. Berman's team observes that models are "breaking the benchmarks way faster than anybody could imagine." The implication is that the window to build defensibility around AI capabilities is shorter than anyone assumed - which makes the physical intelligence thesis, where the software meets physical constraints, more durable as a moat.