$300M Pre-Seed Round: Not a Bubble, But Rational Pricing for Scarce Talent

The most valuable info in this article is that Andrew Dai secured a pre-seed round at a $300 million valuation solely on the labels of "DeepMind Researcher" and "Visual AI," without any product or even a clear roadmap. Many think this number is crazy, but to me, it's the truest reflection of current AI investment—the market is pricing people, not products.

Valuation Logic: Betting on "Scarcity" Rather Than "Certainty"

"I know this founder." — That's the first reaction many top-tier institutions had when seeing Andrew Dai's deck.

Andrew Dai's background is indeed scarce: Core researcher at DeepMind, involved in AlphaFold and Gemini visual modules, with top papers in Visual AI. There are probably fewer than 50 such people globally. And Visual AI—especially video generation, 3D scene understanding, and embodied intelligence—is becoming the next key battlefield in the LLM race.

A $300M pre-seed valuation is essentially an option pricing for "super talent." Investors aren't betting on what product he makes next year, but what his person + direction will be worth in five years. Compare this: If core members of OpenAI's Sora team started a company, valuations might be even higher. This "talent premium" is very common in early stages of paradigm shifts—in 2015, several DeepMind researchers leaving to do autonomous driving followed a similar path.

But here's a key problem: A $300M pre-seed valuation means subsequent rounds must achieve 10x+ growth for investors to profit. This pressure weighs on the founder from day one. Andrew Dai needs to prove he's not just "good at publishing papers," but "capable of building commercially valuable products."

Business Model: The "Cambrian Explosion" and "Darwinism" of Visual AI

Andrew Dai's direction is "Visual AI," but specifically what? The news doesn't say. This is precisely the biggest risk. Currently, Visual AI has three hot tracks:

• Video Generation (Benchmarking Sora, but needing to solve controllability and cost)
• 3D Content Generation (Underlying demand for gaming, film, XR)
• Embodied Intelligence (Robot vision, requiring hardware coordination)

Business models vary hugely across these directions. Video generation leans toward C2C subscriptions or API calls; 3D generation may target B2B tool subscriptions; embodied intelligence requires deep binding with hardware manufacturers. Whether Andrew Dai can quickly lock down a vertical scenario after funding determines the duration of investor confidence.

I noticed he didn't choose the common "build demo first, then raise funds" path, but raised money directly on "vision." This carries an implicit risk: Investor expectations are set too high, and the team might fall into the temptation of "must build platform-level products," ignoring real PMF (Product-Market Fit). Historically, many star teams died at this stage—they were great at research but bad at product.

Competitive Moat: Talent is the Moat, but Also the Ceiling

Andrew Dai's core moat is himself. But the talent moat is bidirectional: He can attract top talent, but can also be poached by better offers. Team execution is the only way to solidify this moat.

Team execution is key—this phrase is especially important in AI, because tech iterates so fast that architectures from three months ago may already be obsolete.

Competition in Visual AI is heating up rapidly. Google has Veo, OpenAI has Sora, Meta has Make-A-Video, plus startups like Pika, Runway, Stability AI. Where is Andrew Dai's differentiation? The news doesn't mention it, but investors should have asked clearly: Is it a more fundamental model architecture? More efficient training methods? Or exclusive data in a specific vertical?

If it's just "a better video generation model," the moat is thin—a three-month lead could be erased by a big player's version update. Real moats should be: Building user stickiness via data flywheels in a niche field, or forming ecosystem bindings via hardware integration.

Investment Judgment: Would I Invest in This Deal?

If I were Andrew Dai's investor, I'd add a condition to the TS: A product prototype must emerge within 6 months, and user retention must hit a certain threshold. This isn't distrust, but hedging against the common ailment of "star researcher startups"—over-pursuing perfection and delaying launches.

A $300M pre-seed valuation is essentially a bet. Betting that Visual AI will be the next "iPhone moment," and Andrew Dai is the one who can seize the opportunity. The stake is huge, but returns could be huge too. For early investors, the key is: Can you bear this risk? And do you have enough post-investment resources to help him land?

[!tip]

Truly smart investors don't just invest in "people," but in the combination of "person + direction + timing." Andrew Dai's direction and timing are right, but the "person" variable still needs verification.

When I see photos of Andrew Dai, I think: He looks like many AI entrepreneurs I've met—smart, confident, slightly exhausted. This exhaustion is worth noting: Has he been sleeping poorly for months? Or is this his "normal" state?

Leaving a question for you all: If after launch, Andrew Dai's product turns out to have no essential difference from Sora, can his valuation hold up?

Original Link: https://techcrunch.com/2026/07/16/how-a-former-deepmind-researcher-raised-at-a-300m-pre-seed-valuation-before-launching-a-product/