Miles Wang spent years working in the deepest corners of OpenAI — not the part that writes poems or answers emails, but the part where models learn to reason about molecular structures, proteins, and biological mechanisms that conventional science takes decades to unravel. Now Wang is in negotiations to leave that ecosystem and found his own AI-powered drug discovery startup, and the number circulating in Silicon Valley for its initial valuation is around $2 billion. Not a single product on the market. Not one approved drug. Just the promise of what someone can do when they've lived from the inside how large language models evolved into something that is no longer just about text.
TechCrunch reported the story on July 14, 2026, and the figure immediately generated split reactions: skeptics pointed out that $2 billion for a company that doesn't exist yet is exactly the kind of euphoria that historically precedes painful corrections in the tech sector. Optimists, on the other hand, point to something more structural — drug discovery is one of the most expensive, slow, and failure-prone processes in all of human industry, and AI has spent the last two or three years proving it can compress that timeline in ways that once seemed like science fiction.
Why It's Worth That Much Before It's Worth Anything
The $2 billion valuation isn't arbitrary or purely aspirational, though it is aggressive. It responds to a concrete market logic: venture investors in life sciences have been looking for exactly this founder profile — someone with top-tier technical credentials in generative AI who decides to apply them to the pharmaceutical sector. Wang is not a traditional biotech figure who learned to use machine learning tools; he is an AI researcher who understands biology as a domain of application. That distinction matters.
The conventional pipeline to develop a drug takes between 10 and 15 years from initial discovery to regulatory approval, and costs on average more than $2.6 billion per successful molecule — not counting the failures that fund that success. AI is already eroding those numbers on several fronts: protein structure prediction (AlphaFold changed the game), synthesis of molecular candidates, simulation of drug-receptor interactions, and more recently, models that suggest clinical trial strategies. If Wang has access to architectures or techniques developed at OpenAI that accelerate any part of that pipeline, the valuation starts to make sense — even if it still stings to read.
The Broader Context: AI and Life Sciences in 2026
What is happening with Wang is not an isolated event. It is the most recent expression of a trend that has been accelerating for at least 18 months: researchers from the major AI houses — OpenAI, Anthropic, DeepMind, Google — are leaving to found startups in vertical industries where data is scarce, cycles are long, and regulatory barriers are high. Healthcare, materials, energy. The thesis is that foundation models have already reached enough maturity that the real value is no longer in building bigger models but in applying them with deep domain expertise.
In the pharmaceutical sector specifically, companies like Isomorphic Labs (DeepMind's spin-off), Recursion Pharmaceuticals, and Insilico Medicine have spent years trying to prove that AI can discover real drugs. Results are mixed but promising: some candidates have reached clinical phases, and the speed of discovery has genuinely compressed. The problem is that reaching clinical phase is not the same as having an approved drug, and the market sometimes forgets that distinction when enthusiasm runs high.
The Real Bet
What is interesting about Wang's position — if the negotiations succeed — is that he enters the game with an advantage that other AI drug discovery startups didn't have when they launched: the language and reasoning models of 2026 are qualitatively different from those of 2021 or 2022. The capacity for multimodal reasoning, for integrating heterogeneous data (genomics, proteomics, historical clinical trial data, scientific literature) and for generating verifiable hypotheses has improved substantially. If Wang built or contributed to the architectures that make this possible, he could arrive at the market with a real technical edge — not just a well-designed PowerPoint deck.
The risk remains the same as always in biotech: time. A startup can have the best AI in the world and still take eight years to prove that its drug works in humans. And eight years is a long time for the market to maintain patience — or for another competitor to arrive first.
For now, the $2 billion is a bet on a name, a track record, and a promise. In Silicon Valley, that is sometimes enough to get started. Whether it is enough to change medicine is an entirely different conversation.
Original source: TechCrunch, July 14, 2026