OpenAI researcher Miles Wang in talks to launch AI drug discovery startup valued at $2B

The exodus from OpenAI's research ranks continues, and this time it's taking aim at one of biotech's most stubborn bottlenecks. Miles Wang, a researcher who joined the ChatGPT maker in 2024 after leaving Harvard's computer science program, is in advanced talks to raise roughly $200 million at a $2 billion valuation for a new venture applying foundation-model-scale AI to drug discovery. Lightspeed Venture Partners is negotiating to lead the round, according to multiple sources familiar with the discussions.

Wang declined to confirm the funding figures or the company's description but did not deny the departure. Lightspeed did not respond to requests for comment. The deal remains fluid and could still shift.

The New Arms Race in AI Drug Discovery

If the terms hold, Wang's startup would enter a suddenly crowded field at a valuation that signals investor conviction more than proven traction. In the past six months alone, Chai Discovery — co-founded by former OpenAI researcher Josh Meier — closed a $400 million round at a $3.8 billion valuation. Isomorphic Labs, a Google DeepMind spinout, secured a $2.1 billion Series B in May. Both are building large-scale models to predict molecular interactions and identify novel therapeutic candidates.

The capital intensity is striking. Traditional biotech Series A rounds historically topped out at $50–80 million. Now, pre-revenue AI drug discovery platforms are commanding nine- and ten-figure valuations before a single IND filing. The rationale: foundation models trained on biological sequence data, chemical structures, and experimental outcomes could compress the decade-long, $2 billion average cost of bringing a new drug to market. But the technical bar is rising just as fast as the check sizes.

Wang's differentiator, according to sources, may be a focus on drug repurposing — identifying new indications for FDA-approved compounds or rescuing assets that failed in late-stage trials for efficacy reasons unrelated to safety. The regulatory pathway for repurposed drugs is dramatically shorter: Phase I safety data already exists, manufacturing is established, and Phase II/III trials can proceed directly to the new indication. Time-to-revenue collapses from 10–12 years to potentially 3–5. For investors staring at the patent cliff facing Big Pharma — $200 billion in annual revenue exposure through 2030 — that compression matters.

From Foundation Models to Wet Lab Validation

Wang's OpenAI tenure was brief but visible. He co-authored work on evaluating how large language models and multimodal systems can automate scientific discovery workflows — literature review, hypothesis generation, experimental design, and data interpretation. That research agenda aligns with the "AI scientist" paradigm that several labs are now chasing: not just a prediction engine, but an autonomous loop that proposes, tests, and iterates.

The question for Wang's venture — and for Chai, Isomorphic, and the half-dozen other well-capitalized entrants — is whether model performance translates to wet-lab success. Predicting binding affinity in silico is a necessary but insufficient condition. Solubility, toxicity, metabolic stability, blood-brain barrier penetration, and manufacturability all emerge in the physical world. The companies that pair foundation models with high-throughput experimental infrastructure — robotic labs, organoid screening, cryo-EM validation — will separate from the pure-play model shops.

Lightspeed's potential lead position is notable. The firm has historically backed deep-tech and hard-science platforms, including several computational biology plays. Their involvement suggests the round is being framed as a platform bet, not a single-asset biotech wager.

The Talent Vacuum and the Founder Age Curve

Wang's profile — 20s, Harvard dropout, OpenAI alumnus — fits a pattern reasserting itself in venture capital. After a post-2015 correction that favored experienced operators, investors are again writing large checks to technical founders in their early 20s who have worked at the frontier of foundation model development. The logic: the architecture of the next generation of scientific AI is being written now, and the people writing it are concentrated in a handful of labs. Access to that talent pool justifies the premium.

But the talent vacuum cuts both ways. Every major AI drug discovery startup is recruiting from the same shallow pool: OpenAI, DeepMind, Anthropic, Meta FAIR, and a few academic groups. Retention risk is real. Wang is expected to bring several OpenAI colleagues with him — a pattern seen at Chai and Isomorphic. The competitive dynamic increasingly resembles a series of coordinated lab departures rather than individual founder moves.

Valuation Discipline vs. FOMO

The $2 billion pre-launch valuation will draw scrutiny. Chai's $3.8 billion and Isomorphic's $2.1 billion (Series B, not seed) set reference points, but neither has disclosed revenue, partnered assets, or clinical candidates. The market is pricing option value on a category level: AI drug discovery as the next major vertical for foundation models, after coding, legal, and customer support. That category bet assumes regulatory pathways remain navigable, biological data access expands, and model scaling laws hold in the life sciences domain.

None of those assumptions are guaranteed. The FDA has not yet established a review framework for AI-designed drugs. Proprietary biological data — the fuel for differentiated models — remains fragmented across pharma, academia, and CROs. And scaling laws that work for language may not transfer cleanly to molecular property prediction, where data density is lower and noise is higher.

Wang's startup, unnamed and still forming, will need to articulate a credible path from model to molecule to IND faster than its better-capitalized peers. The $200 million target, if achieved, buys roughly 24–30 months of runway for a team of 50–80 — sufficient for model development and early experimental validation, but not for clinical programs. Partnerships with mid-size biopharma, asset acquisitions from stalled pipelines, or a royalty-based revenue model could extend the horizon.

What Comes Next

The round's closure will be the first concrete signal. If Lightspeed leads at or near the reported terms, the category's momentum holds. If terms soften or strategic investors replace pure financial leads, the market may be signaling saturation.

Either way, the OpenAI-to-biotech pipeline is now a recognized channel. Wang follows Meier. Others will follow. The question for the industry is not whether AI reshapes drug discovery — it does — but whether the current capital formation produces durable platform companies or a generation of expensive science projects that stall at the IND gate. The next 18 months will answer that.