Dark Matter Lab can give Berkeley researchers up to $1 million before incorporation, split between capital, compute, legal, and operating resources.
Emmanuel Vallod is a partner and head of venture research at Hivemind Capital, where he invests across AI, crypto, payments, and frontier technology. He has taught at Berkeley for 14 years, worked at BlackRock, built an AI infrastructure startup, and now uses research as a direct pipeline for identifying technical founders and emerging markets.
This episode examines a funding gap most venture firms avoid: researchers who may have breakthrough technology but are still inside university labs, years away from incorporation, and unable to access enough compute, data, or hardware to test whether their work can become a company.
Emmanuel’s counterintuitive claim is that the next major AI companies will be technology-first. Distribution was once the primary moat, but he argues that durable AI businesses will increasingly begin with technical breakthroughs that make large-scale distribution possible. Most startups built as wrappers around foundation models will not have lasting defensibility unless they control difficult-to-obtain data or operate inside workflows requiring deep expertise.
He also explains why timing in venture is less precise than investors admit. The companies that become unicorns and decacorns rarely look fashionable when they begin. They often appear too early, too strange, or technically impractical. The investor’s job is not simply to identify a large market, but to determine whether the research direction is correct, whether the technical team understands what it does not know, and whether supplying resources now could compress several years of progress.
In Today’s Episode We Discuss
- 00:01– Emmanuel Vallod’s Journey from Math to Venture Capital
- 03:04– How Teaching at Berkeley Shapes His Investing
- 05:09– Building an AI Infrastructure Startup Before the AI Boom
- 07:53– The Communication Gap Between Technical Founders and VCs
- 09:30– Why Founders Must Tell Investors the Bad News
- 11:05– From Failed Startup to Venture Capital
- 14:26– Hivemind Capital’s Evolution from Fintech to Frontier Technology
- 17:09– The Case for Data Centers in Space
- 19:16– The Market Potential of Space-Based Computing
- 21:41– Technical Barriers to Space Data Centers
- 23:36– Why Timing Is Venture Capital’s Hardest Problem
- 26:17– Why Future Unicorns Rarely Look Like the Cool Kids
- 28:33– Conviction, Luck, and Investing Too Early
- 29:32– The Origin of Dark Matter Lab
- 31:34– Funding Researchers Before Company Formation
- 33:34– How Compute and Data Bottlenecks Delay Breakthrough Research
- 36:03– How Dark Matter Lab Differs from DARPA and Accelerators
- 39:28– Evaluating Deep-Tech Companies Before They Exist
- 41:35– Identifying the Right Research Direction
Dark Matter Lab is designed around that compression. Researchers can receive funding before forming a company, without Hivemind claiming rights to their intellectual property. Emmanuel describes one team that expected to wait three years for a federal grant and another that needed roughly $500,000 just to create an initial training dataset.
Pull Quotes
“The unicorns or decacorns at a point in time were never the cool kids when they started.”
“Stop being an asshole.”
Follow Emmanuel Vallod on LinkedIn: LinkedIn