Why AI Apps Fail — Deep Tech Needs Wide IP Moats

What we talked about.

Sunny Ray talks with Daniel Gold, founder of Castle Fund, a $20 million seed stage fund investing in deep tech startups with wide intellectual property moats in fields like biotech, energy, and aerospace. Daniel traces his path from studying music and computer science at Yale and Stanford, through early startup failures in online radio during the dotcom crash, to a successful exit when Music Match was acquired by Yahoo. He explains how becoming an IP attorney and spending years in business development at Amazon, PlayStation, and Roku shaped his understanding of licensing, partnerships, and leverage between big and small companies. Daniel also describes his evolution from casual angel investor to venture partner at Asymmetry and co chair of digital media at Band of Angels, discussing his first angel check into Redfin. The conversation turns to why he avoids the crowded AI application layer, arguing that software without strong IP protection is highly disruptable, and why he prefers backing founders building defensible technology in the physical world.

An ex IP attorney turned VC explains why he skips hot AI apps for deep tech startups protected by real patents.

The questions, and the answers.

What is Castle Fund and what are you guys up to?

Castle Fund is a $20 million seed stage fund, my first fund, focused on deep tech with wide IP moats. I'm an ex IP attorney and spent 12 years in partnerships and BD roles doing IP licensing for enterprises. AI is disrupting software so fast people can't keep up, so I prefer strong IP in non software fields like biotech, energy, and aerospace. I had my first close in December, made three investments, and I'm gearing up for a second close.

Walk me back to 18 year old Daniel studying music and computer science at Yale and Stanford. What were you really after?

I grew up playing jazz saxophone and classical piano and wanted to see if I could become a professional musician. I deferred Yale a year to explore music in Seattle and also interned in a genetics lab, which taught me I loved reading about biotech discoveries but hated the day to day lab work. At Yale I started studying computer science and later transferred to Stanford for its music, science, and technology program because I wanted access to a killer CS department during the early dotcom boom.

Have you been tinkering with AI music tools like Suno and Udio?

I played with Suno and Udio about a year ago when they first came out, but I don't have much time now with raising a fund and young kids. I'd mainly use them for vocals since I'm not much of a vocalist myself. The quality is unevenly distributed depending on genre, I noticed it was very strong for something like R&B, similar to how AI adoption itself varies a lot across different users and use cases.

What was it like trying to sell people on paying for streaming music in 2001?

It wasn't easy, but we weren't charging much, mainly for higher quality audio and personalization since full on demand song services didn't exist yet because labels hadn't granted those rights. I built an artist radio product where you could type in three artists and get a custom mix. Napster's pressure eventually loosened the labels up, but competing against free stolen music on Napster was still very hard.

Looking back at the producer, the lawyer, and the BD operator, is there one thread running through all of it?

Inventiveness and creativity is the highest level thread, along with being drawn to small teams trying to do something hard that hasn't been done before. That thread evolved into deeply understanding intellectual property, not just as a legal framework but as an actual business, meaning understanding how you actually make money from IP. That became the common thread running through everything.

You did BD at Amazon, PlayStation, and Roku, twenty years inside the IP distribution and licensing machine. What did you learn that most founders never see?

I learned there are supply chain and partnership strategies for growing something small into something big, whether through community, marketing, or licensing, and that the open source movement taught the media sector a lot about balancing free value with paid offerings. I also learned about leverage between big and small companies, that it's often not obvious what a bigger company actually wants, and that many founders need partnerships and channel deals to cross from early adopters to a mass audience.

When did you make your first angel check, and what was the deal and thinking behind it?

My first angel check was into Redfin. I was the first hire the three founders made, as a product consultant, because I'd worked with one of the founders at Amazon. He let me convert some of my consulting fees into investment in the friends and family round. It wasn't a principled bet, I just thought the founders were smart and operating at the cutting edge with first mover advantage, so I figured they'd figure it out.

Most VCs are pouring into the AI application layer, but you're going the other direction toward atoms, molecules, and orbits. What are they missing?

The AI application layer doesn't insulate you from disruption, there's a whole stack from base models to applications to agentic orchestration, and even leaders at one layer can get disrupted by the next hot app a year later. Unless you're backed by massive multi stage funds that can keep a company alive through a slowdown, that growth can evaporate fast. I want IP so strong that even a competitor would rather acquire the company than fight it.