Andrew Dai spent more than a decade at Google, including research that later fed into ChatGPT. Then he walked out, started a company called Elorian, and raised a $55 million seed round at a $300 million valuation. There’s no product. According to TechCrunch AI, which broke down the raise on its Build Mode podcast with Startup Battlefield lead Isabelle Johannessen, Dai pulled this off just months after leaving DeepMind.
What stands out here isn’t the dollar figure. It’s the ratio.
What’s the actual deal?
Dai raised $55 million on a $300 million valuation. That’s roughly 18% dilution on a company with research talent, a thesis, and not much else you can log into.
TechCrunch AI notes the comparison that matters: this is a more aggressive valuation-to-capital ratio than Thinking Machines, which raised one of the largest rounds in U.S. history. Investors on the cap table include Nvidia and Menlo Ventures.
For context, a $300 million pre-product valuation was a fantasy number three years ago. Now it’s a Thursday.
Why visual AI?
Because Dai thinks it’s the gap nobody closed while everyone else raced to build better coders.
“You have models that are doing really great at math, really great at new physics ideas, and of course coding is very popular now … But one area where progress has been extremely uneven is visual understanding and visual reasoning,” Dai told TechCrunch AI. “At Elorian, we want to build models that will advance us toward visual AGI.”
He’s not wrong about the unevenness. Frontier models can write a working compiler and still fumble a chart, misread a diagram, or lose track of objects across frames. Visual reasoning, meaning actually understanding what’s happening in an image or video rather than captioning it, has lagged behind text reasoning by a wide margin. That’s the wedge.
Why turn down a higher valuation?
This is the part founders should read twice. Dai says he had higher offers on the table and passed.
He picked Nvidia and Menlo Ventures instead, betting that investors who understand what building frontier AI actually costs are worth more than a bigger number on a term sheet. Nvidia’s involvement isn’t just capital either. When your bottleneck is compute, having the compute company invested changes what’s possible.
The logic: a higher price today sets a bar you have to clear at the next round. A strategic partner helps you clear it.
How do you sell a technical vision to non-technical money?
Dai’s answer, per the episode, comes down to translation. He refined a deeply technical research direction into a story investors could hold in their heads without a PhD. Not by dumbing it down. By finding the version that’s true and also legible.
Other takeaways he covers:
- Speed is the moat right now. In a field where the frontier moves monthly, shipping fast beats planning well.
- Recruiting from Big Tech is winnable. Researchers leave for scope and ownership, not just equity.
- Jargon loses rooms. If the investor can’t repeat your pitch to their partners, you don’t have a pitch.
Why this matters
This raise is a signal about where capital thinks the next unlock lives. Text and code are crowded, expensive, and dominated by labs with billions in compute. Visual reasoning is comparatively wide open, and it’s where the money is starting to move.
It’s also a market read. Investors are now underwriting people and theses, not traction. A DeepMind pedigree plus a credible frontier thesis is currently worth $300 million pre-product. That’s either efficient pricing of scarce talent or a very expensive lesson in the making. Probably both, depending on the company.
For practitioners: if your work touches document parsing, video analysis, robotics perception, or anything where models need to actually see rather than describe, expect that stack to get better fast. A lot of funded teams are now pointed at exactly that problem.
Elorian still has to ship something. The interesting part starts when it does. The full conversation with Dai is available at the original source.