Resolving Mismatched Time Horizons with Compound’s Michael Dempsey
Michael Dempsey is a managing partner at Compound, a thesis-driven research-centric venture capital firm investing in deeply technical and scientific areas.
Michael Dempsey is a managing partner at Compound, a thesis-driven research-centric venture capital firm investing in deeply technical and scientific areas.
What led you to Compound?
I grew up in the Boston area and went to school in New York. Out of school, I worked at a hedge fund doing both public and private market investing. In 2012, I went to my boss with a long memo proposing we should do early- stage startup investing to better understand the competitive landscape. He asked how much a seed investment cost. I said $250,000 to $500,000, and he said, “Don’t let it impact your day job.”He was one of those bosses who believed in pushing people to pursue their passions.
I made a few early investments and fell in love with it, but figured I should understand how startups actually work. I joined CB Insights pretty early on and helped build the data and research team. At that time, my investing approach was shaped by the hedge fund mentality: generate novel ideas, pressure test them, then invest before anyone else figures it out. It was strange to see how venture worked differently. Everyone stands around waiting to see who jumps in first.
I focused on what we called frontier tech: drones, robotics, VR, space, AI and Bitcoin. I eventually joined what is now Compound in 2016 as we were raising our first fund. Today we have a little north of $200 million under management across several small funds doing thesis-driven, research-centric investing in deeply technical and scientific areas.
What is Compound’s investment thesis?
We try to figure out the futures we believe in and where we think the world is going. Every time we raise a fund, which we invest over four to five years, we identify core areas we think will matter most over the next decade with asymmetric upside. Then we have adjacencies that might be early but interesting.
We take a bottoms-up view. Where do we think the world is going? How does technology push society? How does society push technology? What types of companies accrue value within these areas? We’re best at understanding mismatches in time horizons between what’s technologically possible and what other people think is possible.
Today our core areas are machine learning and automation (including robotics), biotechnology and healthcare, and crypto. We also have adjacencies like energy and materials science. Healthcare is interesting because we did a lot early on, then pulled back, but now we think it will boomerang back with AI in healthcare and societal shifts toward patient-directed medicine.
What causes these mismatches in time horizons that create investment opportunities?
Human nature makes it hard to model change. People get set in their ways of thinking about technological change and company building. There’s the Bill Gates quote about overestimating what you can do in a year but underestimating what you can do in 10 years.
In 2016, the opportunity was understanding things others didn’t from a technological perspective. We could look at low-quality AI generated images from the University of Tokyo and see they were getting 4x better every 12 months. If that continued, it was clear to us that generative AI would be huge. That was a relatively easy extrapolation by understanding scaling laws, compute and data.
Now it’s flipped. People assume everything will be like AI, happening quickly with massive impact. We see this in robotics. Our job now is understanding second and third-order effects and having precise views on value accrual. Context windows are a good example. Companies built complex routing tools assuming small context windows would persist, but they expanded dramatically, making many of those companies obsolete within 12 months.
How do you identify technologies ready for commercialization versus those that should stay in the lab?
We set up Google Scholar citation alerts on interesting research to track velocity of citations. It signals whether ideas are gaining traction or suffering from reproducibility issues. The key question is if you bring a breakthrough to market, what’s the durability of capturing value? People overrate this by thinking cool technology automatically captures value. The biggest lesson for deep tech founders is that the world isn’t a meritocracy. You can build amazing technology but buyers might not care because their career timeline doesn’t align with your deployment timeline, or it’s not a burning enough pain point.
Building organizations that can go from scientific breakthroughs to market is incredibly hard. OpenAI succeeded partly because they had great technical talent but also recruited incredible product talent at the right inflection point. In deep tech, talent is supply-constrained. There aren’t many people who can do what you’re doing.
What patterns do you see in successful deep tech founders?
We see two types. First are founders with precise views of the future seven years out who know exactly how to sequence getting there. They’re excellent at making technical milestones legible to investors and nail timing. Being off-path for six to nine months can be death for these companies. Alex Kendall at Wayve exemplifies this. He had a vision for end-to-end learning applied to self-driving cars and built one of the best embodied intelligence companies over eight years when everyone thought Waymo and Cruise had already won.
The second type understands what their organization excels at and pushes that repeatedly through the “maximum ideas” framework. They know their long-term vision but not the exact path, so they ship the maximum number of experiments to find what works. Cristobal at Runway talks about this. They knew creatives needed AI tools but didn’t know if it would be for imaging, VFX, or video. So they shipped constantly and doubled down on what hit. Companies die when they sit in the middle. They’re not moving quickly but also not being precise. Talent and investors struggle to understand what they’re underwriting.
What’s your contrarian view on deep tech?
I’m relatively anti-humanoid robots. Everyone loves them, but I think technologists hold a strong hammer and look for nails. People misunderstand robotics by thinking about replacing humans and pricing tasks by hourly wages. Instead, they should price human life. Humanoids make sense when you price human life in military or industrial settings where death has consequences beyond hourly wages. For labor arbitrage, purpose-built machines are far better suited.
I also believe most primary care should be done through AI and humans should take health into their own hands at much higher rates. And all investment firms decay rather than compound, which is why we constantly fight against that decay by updating our priors and relitigating our beliefs.
How do you define deep tech?
Deep tech is trying to solve a problem that a subset of people think is largely impossible, or rather largely impossible on the prescribed time horizon.
HAUS specializes in public relations and creative services for deep tech startups.
