An OpenAI Alum Eyes Drug Repurposing as Next AI-Bio Frontier

The Repurposing Bet

Miles Wang spent two years at OpenAI evaluating how large models could speed up biological and scientific workflows. Now he's stepping out to build a company around one of the discipline's most pragmatic problems: finding new therapeutic uses for drugs that already exist.

Wang is pursuing roughly $200 million in capital at a $2 billion pre-money valuation, with Silicon Valley firm Lightspeed in discussions to anchor the round, according to people familiar with the negotiations. Several former OpenAI colleagues are expected to join him. Wang disputed the funding terms and company description but declined to offer alternative figures.

At DailyTechWire, we've tracked a steady stream of AI-biology exits from frontier labs over the past eighteen months, and the pattern is instructive. Founders are no longer pitching general-purpose protein-folding engines; they're carving out narrow, capital-efficient wedges where a model can demonstrate clinical or commercial traction inside thirty-six months. Drug repurposing, sometimes called repositioning, fits that brief neatly. An FDA-approved molecule has already cleared Phase I safety hurdles; the question is whether it can treat a different indication. That path can shave five to eight years off a traditional development timeline and tens of millions of dollars in preclinical spend.

Why Repositioning Appeals to Both Founders and Funders

The thesis rests on two pillars. First, regulatory agencies have already blessed the safety profile of thousands of compounds. Second, modern transformer architectures trained on biomedical corpora, protein interaction data, and clinical trial registries can surface non-obvious connections between a drug's mechanism and an unrelated disease pathway. If the model identifies a credible hypothesis, a startup can move straight into Phase II efficacy trials, bypassing the long, expensive march through toxicology and early-stage human studies.

Investors are warming to the model. Chai Discovery announced a $400 million raise at a $3.8 billion valuation this week; the company builds foundation models to predict molecular interactions. Co-founder Josh Meier also spent time at OpenAI before launching Chai. Google DeepMind's Isomorphic Labs, led by AlphaFold architect Demis Hassabis, closed a $2.1 billion Series B in May. The capital is flowing to teams that can demonstrate a credible path from in-silico prediction to in-human validation.

Wang's background suggests he understands both the model-building side and the constraints of wet-lab biology. He joined OpenAI in 2024 after leaving Harvard, where he was pursuing a computer science degree. During his tenure he co-authored papers examining how AI systems can automate portions of the scientific method, work that sits at the intersection of reasoning models and domain-specific tool use. The leap from automating discovery to commercializing it is shorter than it appears; the hard part is choosing a wedge narrow enough to validate quickly but broad enough to support venture scale.

The Competitive Landscape Is Dense but Segmented

AI-driven drug discovery is no longer a novelty category. Dozens of startups are training models on everything from single-cell RNA sequencing to electron cryomicroscopy images. What differentiates them is the specific bottleneck they attack. Some focus on de novo small-molecule design. Others optimize antibody sequences or predict off-target effects. A smaller cohort, which Wang's venture appears to join, concentrates on repositioning.

The advantage of repositioning is speed; the disadvantage is market size. A repurposed drug rarely commands the exclusivity or pricing power of a novel molecular entity. Regulatory pathways vary by jurisdiction, and in some cases a company must still run full Phase III trials if the new indication is sufficiently different. But for a young company with limited runway, the ability to generate clinical data within two to three years can be the difference between a follow-on round and a down round.

We've observed that venture firms are increasingly willing to fund college dropouts again, a sentiment that had cooled after the high-profile implosions of the 2010s. Wang fits that archetype: he left Harvard to join OpenAI, and he's now raising at a unicorn-plus valuation before his company has a public name. The bet is less on pedigree than on the velocity of learning inside a frontier lab and the strength of the founding team's network.

What Remains Uncertain

Details about Wang's startup remain scarce. The company has no public website, no announced partnerships with pharmaceutical incumbents, and no disclosed data sets or model architecture. It's unclear whether the team intends to operate as a platform licensing models to biopharma or as a vertically integrated drug developer that will carry candidates through trials and seek its own regulatory approvals.

Platform models generate revenue faster but cap upside; vertical integration offers the chance to capture the full value of a successful drug but requires significantly more capital and operational complexity. Most AI-bio startups eventually choose a hybrid: they license the platform to partners while advancing one or two wholly owned programs to demonstrate clinical proof of concept.

Another open question is how Wang's models will handle the messy realities of polypharmacology. Many drugs work through multiple mechanisms, and predicting which secondary target will drive efficacy in a new indication is not a clean computational problem. The best-performing repositioning candidates often emerge from serendipity, clinical observation, or retrospective analysis of patient data, not purely from in-silico screening. A model that can synthesize those diverse signals, including real-world evidence from electronic health records, would represent a meaningful advance.

The Broader Shift in AI-Bio Capital Allocation

The funding environment for AI-biology companies has bifurcated. Early-stage rounds remain competitive, especially for teams with pedigree from OpenAI, DeepMind, or Anthropic. But Series B and beyond now require proof that the model can generate a falsifiable hypothesis, that a wet lab can test it, and that the result moves a molecule closer to the clinic. The era of raising on architectural elegance alone is over.

Wang's reported valuation, $2 billion pre-money, prices in significant future execution risk. For context, Chai Discovery raised at $3.8 billion post-money after two years of operation and multiple disclosed partnerships. Isomorphic Labs, backed by Alphabet, had the luxury of a corporate balance sheet and a Nobel-adjacent founder. Wang will need to move quickly to justify the entry price, likely by announcing a flagship program, a pharma collaboration, or early clinical data within the next twelve to eighteen months.

The talent migration from OpenAI into vertical applications continues to accelerate. In the past year alone, researchers have left to build companies in robotics, materials science, climate modeling, and now drug repositioning. The pattern reflects a broader truth: foundation models are becoming commoditized infrastructure, and the value is shifting to domain-specific fine-tuning, proprietary data, and the ability to navigate regulatory and operational complexity in the physical world.

What Comes Next

If Wang's venture closes the round at the reported terms, it will join a cohort of AI-bio unicorns that must now prove their models can survive contact with biology. The next twelve months will reveal whether drug repositioning, as a category, can support venture-scale outcomes or whether it remains a useful but ultimately niche application of AI in life sciences.

For now, the bet is that the combination of OpenAI-grade modeling talent, a pragmatic go-to-market wedge, and a capital base sufficient to fund multiple clinical programs will be enough to carve out a defensible position in a crowded market. The data will tell.