This Guy Left DeepMind And Raised $55 Million Before He Had A Product

Andrew Dai's entire fundraising story sounds almost fake if you say it out loud. Leaves Google DeepMind. Starts a company. Raises $55 million at a $300 million valuation. Months later. No product yet.

For comparison, that's actually a more aggressive valuation-to-money ratio than Mira Murati's Thinking Machines Lab, and that round was one of the biggest in U.S. startup history. So Dai's not just doing well, he's doing well by a pretty extreme benchmark.

He talked through the whole thing on TechCrunch's Build Mode podcast, and honestly, the story's interesting beyond just "wow big numbers."

Why did he bet on visual AI specifically?

Dai spent over a decade at the center of some of the most important AI research happening anywhere, including work that ended up feeding into how ChatGPT was built. So when he decided to leave and start his own thing, he wasn't picking randomly.

His reasoning is pretty simple once you hear it: AI's gotten really good at math, at generating new physics ideas, and obviously, coding's having a moment right now. But visual understanding, actually reasoning about images the way models reason about text or numbers has lagged way behind everything else.

That gap is what his new company, Elorian, is going after. His stated goal isn't modest either; he's talking about building toward what he calls "visual AGI."

He turned down bigger money

Here's the part that actually surprised me. Dai said he had offers on the table with higher valuations than what he ended up taking. He picked Nvidia and Menlo Ventures instead, specifically because of what those investors bring beyond just the check.

His logic was that investors who actually understand what it takes to build frontier AI are more valuable long-term than just squeezing out the highest number on a valuation. That's a genuinely unusual thing to prioritize when you could just take the bigger offer and brag about a higher number. Most founders don't get offered that choice in the first place, and the ones who do don't always turn down the money.

Turning something technical into something investors actually get

A big chunk of the conversation is about how he actually pitched this. Visual AGI is not an easy concept to explain to someone who isn't deep in AI research. Dai talked about how he had to strip his pitch down from something highly technical into a story that people without a research background could actually follow and get excited about.

That's a skill a lot of technically brilliant founders struggle with being able to build the thing versus being able to explain why the thing matters to someone writing a check.

The other stuff worth knowing

The episode also gets into some genuinely useful, practical territory for anyone paying attention to how AI startups actually operate right now:

• What top VC firms are actually looking for when they evaluate frontier AI bets

• How to recruit researchers away from places like Google or DeepMind, which is its own puzzle given how well those companies pay and how much compute they can offer

• Why speed has become one of the single biggest advantages you can have in AI right now not just being right, but being fast

• How do you even build something durable and defensible when the underlying technology is changing every few months?

It's part of a broader season of Build Mode focused specifically on fundraising, talking to people who've raised massive pre-seed rounds, bootstrapped from nothing, gone public, all of it. Dai's episode fits right into that theme: a founder who managed to raise serious money before he even had something to show, purely on the strength of his background and how well he could communicate the vision.