Superworth

AI went from code to math to biology, and prices pay for other things

Coding already has a loser, priced like one. The biology names rose on a spending recovery, and the trial runner AI could shrink trades at its median.

A thesis by Superworth

Helped

  • GOOGOwns DeepMind and Isomorphic, behind the math and biology milestones. Fwd PE 26.2 near its 25.0 ten-year median.
  • NVDASells the compute every field runs on. Fwd PE 19.1, cheaper than 96% of its ten years, adjusted EPS seen up 71%.
  • AVGOCustom AI chips. Fwd PE 20.9 against a ten-year median of 16.8.
  • SNPSChip design software. Fwd PE 24.2, cheaper than 81% of its ten years, below the price Nvidia paid.
  • CDNSChip design software. Fwd PE 35.6 against a ten-year median of 37.4.
  • LLYIsomorphic partner with an AI lab planned with Nvidia. Fwd PE 28.7 near its 27.3 ten-year median, below its 35.7 three-year one.
  • NVSIsomorphic partner since January 2024. Fwd PE 15.0 against a ten-year median of 14.3.
  • TMOSells lab instruments and also runs trials. Fwd PE 25.3, above 74% of its ten years, up 40% in three months after its July results.
  • DHRLife science tools. Fwd PE 25.3, below its ten-year median of 28.0.
  • ALab instruments. Fwd PE 25.5 near its 24.8 median, with the price at a three-year high.

Exposed

  • IQVRuns drug trials, work AI aims to shorten. Fwd PE 19.8 near its 20.3 median, gross margin lower than 97.5% of its ten years.
  • ACNThe coding precedent. Fwd PE 12.5, cheaper than 97.5% of its ten years.

Worth watching

  • ILMNGene sequencing. Trailing revenue flat at 4.49B over three years while the stock rose 144% in a year.
  • AMATThe next field: AI-found chip materials. Fwd PE 27.5, higher than 96% of its ten years, on the chip cycle.
  1. Code fell first, then math, and biology is next

    Google said in April that AI writes 75% of its new code, up from 50% last fall. Math came next: its Gemini model earned a certified gold at the July 2025 math olympiad. AlphaFold's protein work won a 2024 Nobel. Alphabet did all three, yet trades at 26.2 times the next twelve months, near its ten-year median of 25.0. Its earnings still come from search and cloud. Forward PE history is rebuilt with hindsight.

  2. The compute seller is not priced for a second wave

    Every field AI takes on runs on the same kind of compute. Nvidia committed up to 1B to a drug lab with Lilly in January and put 2B into Synopsys last December. It trades at 19.1 times the next twelve months, cheaper than 96% of its ten years (median 33.9), while consensus sees adjusted EPS up 71%. Broadcom is at 20.9, above its 16.8 median. A low multiple on fast growth says the market doubts how long the spending lasts. Forward PE history is rebuilt with hindsight.

  3. Chip design fell on China, not on agents

    Synopsys lost about a third in one day in September 2025, on results hurt by China sales. Revenue is up 46.3% over the year, with Ansys added since July 2025. It trades at 24.2 times the next twelve months, cheaper than 81% of its ten years (median 31.6), and at 409 sits below the 414.79 a share Nvidia paid. Cadence is at 35.6 against a 37.4 median. AI agents have not moved either price yet. Forward PE history is rebuilt with hindsight.

  4. The drugmakers betting on AI are priced for their drugs

    Lilly and Novartis both signed with Isomorphic, Alphabet's drug design unit, in January 2024, and Lilly agreed a lab with Nvidia this January. Lilly trades at 28.7 times the next twelve months, near its 27.3 ten-year median and below its 35.7 three-year one, on adjusted EPS seen up 29.6%. Novartis is at 15.0 against 14.3. If AI cuts the cost of finding drugs, these prices do not show it yet. Forward PE history is rebuilt with hindsight.

  5. The lab tools rallied on results, not on AI

    AI biology still runs on lab experiments, and Thermo Fisher sells the instruments. It rose 40% in three months, after July results with revenue up 10%, to 25.3 times the next twelve months, higher than 74% of its ten years (median 22.1). Agilent is at 25.5 (median 24.8), Danaher at 25.3 (median 28.0). The rally tracks a spending recovery, so AI biology is not what these prices pay for yet. Forward PE history is rebuilt with hindsight.

  6. Coding's loser is priced as one. Biology's is not

    Drugmakers pay IQVIA to run trials, work AI promises to shorten. OpenAI launched a drug research model in April, yet IQVIA's revenue growth rose from 2.8% to 8.2% since late 2024. It trades at 19.8 times the next twelve months, near its 20.3 median. Accenture, the coding precedent, is at 12.5, cheaper than 97.5% of its ten years. Watch IQVIA's 26.2% gross margin, lower than 97.5% of its ten years. Forward PE history is rebuilt with hindsight.

  7. What would prove it wrong: biology has not reached revenue

    If AI were solving biology, the makers of biological data should sell more. Illumina's trailing revenue is 4.49B, near the 4.46B of three years ago, partly because China banned its sequencers in 2025 and university labs lost funding. Yet the stock rose 144% in a year to 43.4 times forward earnings, on revenue seen up 5.5%. Rentosertib, the first generative AI drug in Phase 3, dosed its first patient on 9 September in a 52-week trial.

  8. What comes next: materials, at a price set by the chip cycle

    After math and biology, labs aim at materials. DeepMind plans a UK robot lab in 2026 to find superconductors and chip materials. Applied Materials trades at 27.5 times the next twelve months, higher than 96% of its ten years (median 15.2), down from 46.1 in June. That is the chip cycle: adjusted EPS is seen up 57.5% and revenue up 41.2%. Materials discovery is not in the price. Watch Nvidia's 18 November results. Forward PE history is rebuilt with hindsight.

1 of 8

What has happened lately

Headlines link to the original source.

Brought to you by Superworth

An idea, tested against what the prices already assume.

A thesis tests an idea against prices. It is not investment advice.