Qnite Labs

The field, with its losses

AI drug discovery has raised billions and produced real progress. It has also produced write-downs, layoffs and failed trials. This page sets out both, with sources, and the risks we carry too.

Public figures as of October 2026.

Most drugs fail, at great cost

Of every 100 drugs that enter human trials, about 8 are approved.28 Each approved drug costs the largest pharma companies $2.67B on average, once the failures are paid for.31 This is the business everyone here is in, including us.

  1. 100Phase I
  2. 52Phase II−48
  3. 15Phase III−37
  4. 8.7Filed−6.3
  5. 7.9Approved−0.8
Drugs remaining out of 100 that enter Phase I; the bottom row shows how many drop out at each step. Phase II is where most are lost. Computed from BIO’s 2011–2020 phase success rates.28

Even the leaders lose money

Revenue and net result in 2025, in US$ millions. Recursion lost about $8.60 for every dollar it earned.1 XtalPi is the exception: its first profitable year.5

  • Recursion1
    Revenue$74.7M
    Net loss−$644.8M
  • Schrödinger2
    Revenue$255.9M
    Net loss−$103.3M
  • Insilico Medicine3,4
    Revenue$56.2M
    Net loss−$43.8M*
  • XtalPi5
    Revenue$111.5M
    Net profit+$18.7M

* Insilico’s adjusted loss. Its reported loss was $352.3M, mostly a one-off, non-cash charge on shares converted at its IPO.4 XtalPi converted from RMB at 7.2 per US$.

Headline value is not cash

The share of each announced deal paid at signing. The rest depends on milestones that most programmes never reach.

$0.5B of $10B

Recursion had received just over $500M from partners, against more than $10B in possible milestones.12

  • Insilico Medicinewith SK Biopharmaceuticals, 202610
    0.7%≤ $18M of $2.5B
  • XtalPiwith DoveTree, 20259
    0.9%$51M of $5.99B
  • Exscientiawith Sanofi, 20228
    1.9%$100M of $5.2B
  • Isomorphic Labswith Eli Lilly, 20246
    2.6%$45M of $1.7B
  • Isomorphic Labswith Novartis, 20246
    3.0%$37.5M of $1.2B
  • All preclinical deals, 2024industry benchmark11
    4.9%$1.3B of $26.8B

Percentages are our arithmetic: upfront ÷ announced total value.

Where it went wrong

Since 2023, many of the best-known AI drug companies have cut staff, dropped programmes or lost their listing. These are serious teams with serious funding.

$688Mall-stock sale
Put the first AI-designed drug into human trials. Then cut its pipeline, fired its CEO and laid off about a quarter of staff, before an all-stock sale worth about $688M.13,14,15
Privateleft Euronext Amsterdam
Once one of Europe’s best-known AI drug companies. Left Euronext Amsterdam after a merger with Osaka Holdings took it private.16
−20%of staff laid off
Cut programmes in NF2, a vascular brain disorder and C. difficile, and laid off a fifth of its staff.17
~90%of staff cut
Its AI-found ALS drug did not help patients: a marker of nerve damage went up, not down. Relaunched as a data company.18
114patients, no clear benefit
OPL-0401, a ROCK inhibitor, showed no clear benefit over placebo in 114 people with diabetic retinopathy.19
−22%about 65 roles
About 65 roles cut to make its cash last into 2027, citing a difficult market.20
−30%of staff
Laid off staff months after a Sanofi deal, then moved from partnerships to developing its own drug.21,22

Our reading

Two patterns repeat. The trial failures were mostly biology, the wrong target or the wrong patients, rather than poor chemistry. And the cuts came after large own pipelines were funded before the platform had proven itself. Target choice is the step we work on, and it is where we can be wrong too.

And what has worked

How the leaders make money

CompanyHow it earns2025What it shows
Insilico MedicineSoftware subscriptions, and licensing its own drug candidates$56.2M revenueSoftware used by 13 of the top 20 pharma companies3
SchrödingerSoftware, plus discovery collaborations$255.9M revenueSoftware is 78% of revenue; still loss-making2
RecursionIts own pipeline, funded by partnerships$74.7M revenueMore than $500M received from partners since founding1,12
XtalPiDiscovery services, robotics and partnered assetsRMB 802.6M revenueFirst profitable year5
Isomorphic LabsBig-pharma partnerships and its own pipelinePrivate$82.5M upfront from Lilly and Novartis; ~$2.7B raised6,7

Our risks, plainly

We are not the answer to the industry’s losses. We are a bet that better target choices lower them. Here is what could go wrong.

  1. Our target calls can be wrong

    Genetic evidence makes a drug about 2.6 times more likely to be approved.29 Starting from 7.9%,28 that is still only about one in five. Even if we do our job well, most programmes we support could fail.

  2. Proof takes years

    A target chosen today reaches a clinical answer in 5 to 10 years. Until then our evidence is retrospective tests, lab results and client decisions. AI-designed drugs still pass Phase II only about 40% of the time, the industry norm.27

  3. Pharma pays little upfront

    Most deal value arrives late or never, as the chart above shows. We plan around cash received, not headlines.

  4. Models are becoming free

    Open models and big-pharma offers such as Lilly’s TuneLab30 push model prices towards zero, and one competitor has just raised $2.1B.7 We can’t win on model size, only on decisions that hold up.

  5. We start without our own lab data

    Early on we rely on public data and partners’ experiments. Building proprietary data takes time and money.

  6. The industry itself is under pressure

    Forecast return on pharma R&D is 2.9% once GLP-1 drugs are excluded.31 Budgets for outside discovery work can shrink fast.

We may fail. This business fails often, and good teams with good science have failed in it. We are building so that a wrong call costs a client weeks and a report, not years and a clinical trial.

How we plan to make money

Paid work first, upside later. The companies above that ran out of room mostly spent on their own pipelines before their platform had proven itself.

  1. Now

    Paid target dossiers

    Fixed-price go / no-go reports on a target, then designed molecules for the ones that pass. Cash early; proof with each client.

  2. Next

    Programme subscriptions

    Annual work on a client’s programme, with milestones or royalties on molecules we help design.

  3. Later

    Selected own assets

    Only with a track record, and licensed out early. We will not build a cash-burning pipeline first.

Questions about these numbers?

We are happy to walk investors through the sources and our assumptions.

  1. Recursion reports Q4 and full-year 2025 financial results (2026).
  2. Schrödinger reports fourth quarter and full-year 2025 financial results. BioSpace (2026).
  3. Insilico Medicine announces 2025 annual results (2026).
  4. InSilico Medicine: 2025 revenue falls 34.5% to US$56.24 million; net loss US$352.32 million. Tiger Brokers (2026).
  5. XtalPi Holdings announces full-year 2025 annual results. PR Newswire (2026).
  6. Lilly, Novartis sign AI partnerships with Alphabet’s Isomorphic. BioSpace (2024).
  7. Isomorphic Labs raises $2.1 billion. BioPharm International (2026).
  8. Sanofi and Exscientia research collaboration (press release, 2022).
  9. XtalPi and DoveTree announce $6 billion AI drug discovery collaboration. Pharmaceutical Executive (2025).
  10. Insilico Medicine and SK Biopharmaceuticals collaboration worth up to $2.5 billion. BioSpace (2026).
  11. DealForma. Global healthcare and life sciences partnerships 2024.
  12. Recursion Pharmaceuticals, Form 10-Q for Q3 2025. SEC.
  13. Exscientia claims world first as AI-created drug enters clinic. pharmaphorum (2020).
  14. After a tough year, Exscientia folds into Recursion. Fierce Biotech (2024).
  15. Recursion acquires Exscientia in $688 million all-stock deal. BiopharmaTrend (2024).
  16. BenevolentAI: result of extraordinary general meeting and delisting. Euronext (2025).
  17. Recursion lays off 20% of staff in wake of pipeline cutbacks. Fierce Biotech (2025).
  18. Verge Genomics relaunches as Verge Labs after ALS drug failure. BioPharma Dive (2026).
  19. Valo Health drops ROCK inhibitor after Phase 2 diabetic retinopathy setback. Patsnap Synapse (2025).
  20. insitro cuts 22% of workforce, extending runway into 2027. BioSpace (2025).
  21. Just months after Sanofi AI deal, Atomwise laid off 30% of staffers. Endpoints News (2023).
  22. “Faster isn’t better”: Atomwise moves from partnerships to pipeline. Endpoints News (2023).
  23. Xu Z. et al. A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a phase 2a trial. Nat Med (2025).
  24. Insilico initiates Phase III clinical trial of rentosertib (2026).
  25. Takeda’s TYK2 psoriasis drug zasocitinib succeeds in Phase 3. BioPharma Dive (2025).
  26. Merck inks $349M biobuck deal with Variational AI. Fierce Biotech (2025).
  27. Jayatunga M.K.P. et al. How successful are AI-discovered drugs in clinical trials? Drug Discov Today (2024).
  28. BIO, Informa Pharma Intelligence & QLS Advisors. Clinical Development Success Rates 2011–2020 (2021).
  29. Minikel E.V. et al. Refining the impact of genetic evidence on clinical success. Nature (2024).
  30. Lilly launches AI drug discovery platform for biotech partners (TuneLab). DCAT Value Chain Insights (2025).
  31. Deloitte. Measuring the return from pharmaceutical innovation (2026).

Company names and icons are trademarks of their owners, shown only to identify them. No company listed here endorses or is affiliated with Qnite Labs.