VC & Startups/Nov 14 edition/Reporting & analysis
ValuationsM&AVenture funds

ValuationsMarks, multiples & secondaries

a16z argues AI drug discovery could earn big pharma software-style valuations

Andreessen Horowitz contends that if AI makes drug discovery repeatable, investors could value pharma companies' discovery engines like software platforms, not just their approved drugs. Frontier AI labs are moving into drug programs, but clinical proof that AI raises success rates is still missing.

From the a16z article of November 13, 2026 and cited sources: Investors now value pharma mainly on approved drugs, and pipeline assets count for little. a16z argues that cutting the cost per approved drug, about $2B in present value by its estimate, would change that math before it changes… [1] [4]

SOURCES[1] Big Pharma Can Rival Big Software a16z.news[4] Fierce Biotech: Lilly–a16z $500M fund fiercebiotech.com

AI-assisted, not human-edited. Methodology Corrections Who owns this site

KEY FACTS
What happened
Investors now value pharma mainly on approved drugs, and pipeline assets count for little. a16z argues that cutting the cost per approved drug, about $2B in present value by its estimate, would change that math before it changes… [1] [4]
Original source
a16z article, independent perspective
Brief by
AT Pulse research desk AI-assisted; publisher and contact on the desk page
When
November 13, 2026 article published
Evidence
14 sources across 14 sites
70/100StrongHow it’s measured

Substack · 109 likes · 4 comments

Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.

Illustration from a16z: a16z argues AI drug discovery could earn big pharma software-style valuations
Image: a16z — Original article ↗
THE CORE IDEAS4 TAKEAWAYS
01

Investors now value pharma mainly on approved drugs, and pipeline assets count for little. a16z argues that cutting the cost per approved drug, about $2B in present value by its estimate, would change that math before it changes valuation multiples. The firm is an interested party: it has backed more than 75 life-sciences startups and manages a fund of up to $500M that Eli Lilly fully finances. [1] [4]

02

Frontier AI labs are moving from selling tools to running drug programs themselves. OpenAI released GPT-Rosalind, a life-sciences reasoning model available through limited access, then bought Orchestra Bio, whose agent layer coordinates drug programs, on undisclosed terms. Anthropic paired Claude Science with its own wet lab and bought Coefficient Bio for about $400M in stock. Anthropic has not disclosed a drug candidate. [5] [6] [8] [9] [11] [7]

03

Headline values in AI-pharma deals depend mostly on future milestones. In Isomorphic's 2024 deals, Lilly paid $45M upfront against up to $1.7B in milestones, and Novartis paid $37.5M upfront against up to $1.2B. [3]

04

Data is the main bottleneck. The Protein Data Bank holds roughly 300K experimentally determined structures, and biological feedback can take years. That favors setups where lab results flow straight back into model training, and the essay expects more demand for contract research organizations and lab automation. [1]

WHY IT MATTERS

Documented: frontier labs now buy biotech companies and run their own labs, and pharma AI deals lean on contingent milestones.

Read the full assessment

Implied: startups selling regulatory or trial-management agents face direct overlap, while wet-lab capacity and proprietary experimental data may prove more defensible.

Executive brief

Andreessen Horowitz (a16z) argues that AI could change how drug companies are valued. Today investors pay mostly for a company's approved drugs. If AI makes drug discovery repeatable, they might also pay for the discovery engine itself, the way they reward software platforms. The authors invest in this area and are not neutral: a16z says it has backed 75+ life-sciences startups and runs a fund with Eli Lilly (a16z). The scale is real. Lilly's Q3 2026 revenue was $24.4B, and tirzepatide (Mounjaro plus U.S. Zepbound) made up about $16B of it (Lilly 8-K). Whether AI improves the odds of a drug succeeding has not yet been shown in clinical trials.

What changed and event timeline

  1. Isomorphic signs with Lilly and Novartis

    Lilly paid $45M upfront with up to $1.7B in milestones. Novartis paid $37.5M upfront with up to $1.2B. Most of the headline value depends on future milestones ().

  2. Lilly–a16z fund launches

    Lilly provides all of the capital, up to $500M, and a16z Bio + Health manages it. AI tools are one stated focus ().

  3. OpenAI launches GPT-Rosalind

    It is a reasoning model tuned for life sciences, available through limited access. Amgen, Moderna and Novo Nordisk are named customers (;).

  4. Anthropic's wet lab revealed

    Reuters reported a Bay Area biology lab, the purchase of Coefficient Bio for about $400M in stock, and a focus on neglected diseases. No drug candidate was disclosed ().

  5. OpenAI acquires Orchestra Bio

    Orchestra's agent layer for coordinating drug programs is being folded into Rosalind. Terms were not disclosed (;).

  6. a16z publishes its thesis

    The essay says bio-AI is at roughly "GPT-3 maturity." It lists pharma partners for Chai, Boltz, BigHat, Nabla and others ().

Capabilities and access

  • GPT-Rosalind (OpenAI): OpenAI reports it beats GPT-5, GPT-5.2 and GPT-5.4 on chemistry, protein understanding, experiment design and tool use. A Codex plugin connects it to 50+ scientific tools.
  • Anthropic: Claude Science software plus an in-house wet lab where Claude directs experiments. No clinical trials are under way (CNBC; Reuters via Yahoo).
Read the full section
  • GPT-Rosalind (OpenAI): OpenAI reports it beats GPT-5, GPT-5.2 and GPT-5.4 on chemistry, protein understanding, experiment design and tool use. A Codex plugin connects it to 50+ scientific tools. Access is through a trusted-access program, worldwide for eligible organizations since Sep 11, 2026 (OpenAI; The Decoder).
  • Anthropic: Claude Science software plus an in-house wet lab where Claude directs experiments. No clinical trials are under way (CNBC; Reuters via Yahoo).

Technical analysis for researchers and developers

  • a16z describes two main kinds of model: (1) LLMs and reasoning models for literature review, experiment design and regulatory writing; (2) diffusion-based models that design or optimize new molecules from scratch.
  • The bottleneck is data. The Protein Data Bank holds about 300K experimentally determined structures.
  • Biology feedback loops are long: whether a design works may not be known for years.
Read the full section
  • a16z describes two main kinds of model: (1) LLMs and reasoning models for literature review, experiment design and regulatory writing; (2) diffusion-based models that design or optimize new molecules from scratch. It says scaling laws "appear to be holding" for molecule generation but cites no benchmark (a16z).
  • The bottleneck is data. The Protein Data Bank holds about 300K experimentally determined structures. That is why the essay expects more demand for contract research organizations (CROs) and lab automation.
  • Biology feedback loops are long: whether a design works may not be known for years. This favors setups where lab results flow straight back into model training, as in Anthropic's lab and full-stack biotechs.

Claims and evidence

  • Lilly GLP-1 "close to $17B" in Q3: the filing confirms Mounjaro at $10.7B and U.S. Zepbound at $5.3B (Lilly 8-K).
  • Partnership values: Isomorphic's $1.2B and $1.7B are potential milestone payments, not cash received (BioSpace).
  • "Anthropic announced its own drug candidates": Reuters reports no candidate or target disease has been disclosed (Reuters via Yahoo).
Read the full section
  • Lilly GLP-1 "close to $17B" in Q3: the filing confirms Mounjaro at $10.7B and U.S. Zepbound at $5.3B (Lilly 8-K). The rest of the gap, and the essay's comparison to Anthropic's quarterly revenue, is not corroborated by the reviewed sources.
  • Partnership values: Isomorphic's $1.2B and $1.7B are potential milestone payments, not cash received (BioSpace). The other partner lists in the essay are a16z's own assertions.
  • "Anthropic announced its own drug candidates": Reuters reports no candidate or target disease has been disclosed (Reuters via Yahoo).

Context and prior work

Eroom's law comes from a 2012 paper in Nature Reviews Drug Discovery. It found that FDA approvals per inflation-adjusted $1B of R&D have halved about every nine years since 1950 (Nature; OECD). Lilly became the first pharma company worth $1T on Nov 21, 2025, driven by GLP-1 drugs (Al Jazeera).

Limitations, safety and contested findings

  • The essay itself admits that biology has little data, is highly complex, and gives slow feedback.
  • The authors have conflicts of interest. a16z names insitro, Formation Bio and BigHat as portfolio companies, and Lilly is its fund partner.
  • Being "near GPT-3" is an analogy, not a measured result.
Read the full section
  • The essay itself admits that biology has little data, is highly complex, and gives slow feedback. It also cites Daphne Koller's point that better molecular design does not help if the biological target is wrong (a16z).
  • The authors have conflicts of interest. a16z names insitro, Formation Bio and BigHat as portfolio companies, and Lilly is its fund partner. Their positive framing of these companies and of Lilly should be read with that in mind.
  • Being "near GPT-3" is an analogy, not a measured result.

Business and practitioner implications

  • Valuation: big pharma trades at 8–20x earnings, and drugs still in development count for little.
  • Competition: frontier labs are moving from selling tools to running programs themselves (Rosalind plus Orchestra; Claude plus a wet lab).
  • Data: a16z sizes the CRO market at $90B. Wet-lab capacity and proprietary experimental data look like the more defensible positions.
Read the full section
  • Valuation: big pharma trades at 8–20x earnings, and drugs still in development count for little. A drop in development cost (now about $2B present value per approved drug, per a16z) would change that math before it changes valuation multiples.
  • Competition: frontier labs are moving from selling tools to running programs themselves (Rosalind plus Orchestra; Claude plus a wet lab). Vertical startups selling agents for regulatory documents or trial management face direct overlap.
  • Data: a16z sizes the CRO market at $90B. Wet-lab capacity and proprietary experimental data look like the more defensible positions.
FOLLOW THE EVIDENCE

The source trail.

Sources (14)
A LITTLE LESS NOISE. A LOT MORE CONTEXT.

Stay curious.
Follow the evidence.

Independent perspectives, the original sources, and room for the questions that don't have easy answers.

How we build the brief
Connect with us

Find us where you already read.