The AI Bubble Is Being Tested. What Does That Mean for Pharma?

Big pharma has made serious AI bets. The real question now is where those bets are changing outcomes.

A needle about to pop a soap bubble holding AI, DNA, lab glassware and medicine icons, against a falling red stock chart

On November 5, the Nasdaq-100 fell 6.4% in a single day. Nvidia fell about 9%, and about $2.5 trillion in Nasdaq-100 market value disappeared. Reuters linked the selloff to fears that the Democratic sweep in the US midterms could slow the build-out of AI data centres [1].

It was not a verdict on AI itself. But it was a reminder that markets are starting to ask harder questions about the returns on massive AI investment. Pharma should be asking similar questions.

Over the last few years, AI has become part of almost every pharma conversation: faster drug discovery, smarter clinical trials, quicker regulatory reviews and sharper commercial decisions. The selloff raises a simple question for the industry: how much of this is delivering real value, and how much still rests on future promises?

The answer depends on where you look. Some uses of AI are already part of everyday work. Some have promising but narrow evidence. And some still have to prove they can change the economics of drug development.

Big pharma’s AI bets: disciplined or frothy?

Eli Lilly

In October 2025, Eli Lilly announced it was building an AI supercomputer with more than 1,000 Nvidia GPUs [2]. In January, the two companies said they plan to invest up to $1 billion over five years in a joint AI lab [3].

Novo Nordisk

Novo Nordisk has signed three major AI partnerships this year, with OpenAI, Amazon Web Services and Anthropic, without disclosing the financial terms [4][5][6].

On paper, that looks like the same kind of AI enthusiasm investors just punished. The numbers tell a calmer story. Lilly spent $13.3 billion on R&D in 2025 [7]. The lab is a joint commitment that covers people, infrastructure and compute, so it is not a like-for-like comparison. But even if Lilly carried the full amount alone, it would come to about $200 million a year, roughly 1.5% of its research budget. Novo’s three deals also cover different needs: AI across the whole business with OpenAI, cloud and drug discovery tools with AWS, and drug discovery work with Anthropic. That points to a company exploring AI across many parts of its operations, not relying on a single use case.

Still, a big investment or a long list of partnerships proves nothing on its own. What matters is whether these bets produce better drug candidates, faster trials or more reliable supply. The sharper bubble risk sits with smaller AI-first biotechs whose valuations rest on platforms that have yet to produce an approved drug. They run on investor patience, and investor patience is exactly what a selloff tests.

Where AI is already delivering

The clearest value today is rarely in discovering the next blockbuster. It is in helping teams work through large volumes of information and repetitive tasks.

Drug safety.

The FDA receives more than 2 million adverse event and medication error reports every year [8]. Its InfoViP tool uses natural language processing to flag likely duplicate reports for safety reviewers to confirm, and in 2023 the agency used it to de-duplicate its entire adverse event database of 28 million reports [9]. Finding duplicates is a narrower job than spotting a new safety risk, but duplicate reports can distort safety analysis, so cleaner data matters [10]. It is unglamorous work, done at scale and checked by experts, which is exactly why it holds up.

2 million+
Reports every year
28 million
Reports de-duplicated

Clinical trial recruitment.

Recruitment delays affect around 80% of trials. AI can help by analysing health records, identifying eligible patients and finding sites with better access to the target population. A 2026 review in the International Journal of Medical Informatics highlighted studies reporting enrolment improvements of up to 65% with certain AI-enabled approaches [11]. That is the top of the range, and results vary widely by setting. But slow recruitment is expensive, and a shorter recruitment period can bring trial results, and potentially launches, closer.

around 80%
Of trials hit by recruitment delays
up to 65%
Enrolment improvements

A more concrete example comes from the FDA itself. In December 2025, it qualified AIM-NASH, an AI tool that helps pathologists score liver biopsies in MASH trials, as its first qualified AI drug development tool. Pathologists still review and accept or reject every score [12]. It solves one defined problem, the variation in how biopsies are scored, and that is a far stronger case than a general claim that AI makes trials faster.

Where the evidence is promising, but still narrow

Regulatory review.

The FDA announced its internal generative AI tool, Elsa, in June 2025, after first deploying it in May, and says more than 70% of its staff have used it voluntarily [13][14]. In May 2026, the agency released Elsa 4.0 and pulled more than 40 internal data systems into a new platform called HALO [15]. For sponsors, that raises the value of submissions that are structured, consistent and backed by clean data, because a machine will often read them first. Adoption is real. Whether it shortens review times is yet to be shown.

70%+
Of FDA staff have used Elsa
40+
Internal data systems in HALO

Patient access.

In the US, prior authorisation can create long delays between a prescription and a patient actually starting treatment. In an American Medical Association survey of 1,000 physicians released in May 2026, 95% said it delays access to care, and 79% said patients abandon treatment because of it [16]. AI tools are now being pitched to complete forms, flag likely denials and keep patients updated on their status [17]. Some of this is AI in the strict sense; some is good workflow automation with an AI label. The difference matters when judging what the technology itself is adding. The potential benefit goes beyond less admin: fewer patients dropping off before treatment, with a direct commercial as well as patient impact. The case will be much stronger once independent results, beyond vendor case studies, are published.

95%
Said it delays access to care
79%
Said patients abandon treatment because of it

Marketing content review.

AI makes content faster to produce, which moves the bottleneck to medical, legal and regulatory review [18]. The bigger opportunity is faster review: checking claims against approved evidence and flagging compliance issues early, with a human still signing off.

Where the promise is still ahead of
the proof

Drug discovery.

This is the most exciting part of the AI story, and the one where expectations need the most care. A 2024 analysis of AI-native biotech pipelines found that AI-discovered molecules had a Phase 1 success rate of around 80–90%. In Phase 2, where a drug first has to show it works, the rate was about 40%, close to the industry average and based on a small sample [19].

80–90%
Phase 1 success rate
~40%
Phase 2 success rate

One of the most closely watched tests is Insilico Medicine’s rentosertib, which started a 52-week Phase 3 trial in about 320 patients with idiopathic pulmonary fibrosis (IPF), with the first patient dosed in September 2026 [20][21]. The trial’s lead investigator has said that, even under favourable conditions, regulatory approval could still be three to four years away [21].

Not every story ends well. In 2023, BEN-2293, an eczema drug from AI drug discovery company BenevolentAI, met its safety goals in a Phase 2a trial but missed its efficacy endpoints [22]. The company stopped investing in it and, as part of a wider restructuring, announced proposed cuts of up to about 180 jobs [23]. More recently, Verge Genomics dropped VRG50635, its AI-discovered ALS drug, after it failed a pre-specified efficacy analysis in an early-stage trial [24]. Both cases point to the same lesson: a promising drug candidate, however it was found, still has to work in people.

Manufacturing.

Lilly and Nvidia plan to apply AI, robotics and digital twins to improve production efficiency and reduce downtime [2][3]. With sales of Lilly’s tirzepatide medicines roughly doubling in 2025 [7], the case for trying is easy to see. But it needs heavy investment, and the benefits on cost, output and reliability still have to be shown.

The forecasting blind spot

For commercial and forecasting teams, AI already speeds up the routine work: cleaning and matching data, refreshing models, spotting anomalies and running scenarios. But a faster analysis is a different thing from a better forecast. Forecasts rest on assumptions about patient starts, persistence, access and competition, and someone still has to understand and own those assumptions.

There is also a less obvious effect. Sometimes, the AI application that has the biggest impact on a forecast is not the one being used by the forecasting team. It is the one improving something else in the patient journey.

  • Faster trial recruitment can move a launch date.
  • Fewer access delays can lift treatment starts.
  • Better adherence can change persistence.

That means AI’s value in forecasting should be judged two ways: how much faster a team can build a forecast, and whether AI elsewhere is changing the patient behaviour the forecast is trying to predict.

Five questions to ask before approving an
AI project

For pharma leaders weighing AI budgets after the selloff, five questions help separate a project worth funding from a demo:

  1. What problem are we solving, and which number will move?Start with the problem, then the technology. “Efficiency” is not a number. Time to first dose, enrolment per site or drop-off rate are.
  2. What is the baseline today?Without a measured starting point, nobody can prove the gain later.
  3. Is the evidence from our data or the vendor’s?A pilot on the company’s own data should come before a multi-year deal.
  4. Will people actually use it?A tool that shines in a demo can fail inside a real workflow. Integration, training and data quality decide whether it sticks.
  5. Who owns the outcome, and when do we stop?Every project needs a business owner, a human check wherever decisions affect patients, and staged funding with a clear point to scale, change or stop.
And one test for the whole portfolio: would this still get funded if AI stocks fell another 30%? If the answer depends on market mood, it is a bet, not a project.

What pharma should take away

The selloff is unlikely to slow AI adoption in pharma. It is more likely to change which projects get funded. The first AI wins that big pharma can report with real numbers will probably come from trial enrolment, safety processing and patient access, well before any AI-discovered drug reaches the market.

AI does not need to discover a blockbuster drug tomorrow to prove its worth. Recruiting patients faster, processing safety cases more efficiently or keeping one more prescribed patient on therapy is already meaningful progress, and it shows up in the numbers.

Which AI tool in your organisation would still get funded if the budget were cut tomorrow?

References
  1. Reuters (via GV Wire) – Blue tsunami wipes out $2.5 trillion from Nasdaq, as investors fret about AI trade (Nov 5, 2026): https://gvwire.com/2026/11/05/blue-tsunami-wipes-out-2-5-trillion-from-nasdaq-as-investors-fret-about-ai-trade/
  2. Eli Lilly press release – Lilly partners with NVIDIA to build the industry’s most powerful AI supercomputer (Oct 28, 2025): https://investor.lilly.com/news-releases/news-release-details/lilly-partners-nvidia-build-industrys-most-powerful-ai
  3. NVIDIA press release – NVIDIA and Lilly announce co-innovation AI lab, up to $1 billion over five years (Jan 12, 2026): https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-and-Lilly-Announce-Co-Innovation-AI-Lab-to-Reinvent-Drug-Discovery-in-the-Age-of-AI/
  4. Novo Nordisk press release – Novo Nordisk and OpenAI partner to transform how medicines are discovered and delivered (Apr 14, 2026): https://www.novonordisk.com/news-and-media/news-and-ir-materials/news-details.html?id=916532
  5. Novo Nordisk press release – Novo Nordisk and AWS enter strategic partnership to accelerate drug discovery through AI (Aug 10, 2026): https://www.novonordisk.com/news-and-media/news-and-ir-materials/news-details.html?id=916594
  6. Novo Nordisk press release – Drug discovery collaboration with Anthropic (Sep 16, 2026): https://www.novonordisk.com/news-and-media/news-and-ir-materials/news-details.html?id=916768
  7. Eli Lilly – Form 10-K for 2025 (R&D expense $13.3 billion; product revenue): https://www.sec.gov/Archives/edgar/data/59478/000005947826000013/lly-20251231.htm
  8. FDA – Office of Surveillance and Epidemiology (evaluates more than 2 million adverse event and medication error reports a year): https://www.fda.gov/about-fda/cder-offices-and-divisions/office-surveillance-and-epidemiology-ose-divisions
  9. FDA – Office of Surveillance and Epidemiology 2023 Annual Report (InfoViP de-duplication of the 28-million-report FAERS database): https://fda.gov/media/180347/download
  10. Kreimeyer et al. – Increased confidence in deduplication of drug safety reports with natural language processing of narratives at the US FDA. Frontiers in Drug Safety and Regulation (2022): https://doi.org/10.3389/fdsfr.2022.918897
  11. Olawade et al. – Artificial intelligence in clinical trials: a comprehensive review. International Journal of Medical Informatics 206 (2026) 106141: https://doi.org/10.1016/j.ijmedinf.2025.106141
  12. FDA – FDA qualifies first AI drug development tool, will be used in MASH clinical trials (Dec 8, 2025): https://www.fda.gov/drugs/drug-safety-and-availability/fda-qualifies-first-ai-drug-development-tool-will-be-used-mash-clinical-trials
  13. FDA press release – FDA launches agency-wide AI tool to optimize performance for the American people (Jun 2, 2025): https://www.fda.gov/news-events/press-announcements/fda-launches-agency-wide-ai-tool-optimize-performance-american-people
  14. FDA press release – FDA expands artificial intelligence capabilities with agentic AI deployment (Dec 1, 2025): https://www.fda.gov/news-events/press-announcements/fda-expands-artificial-intelligence-capabilities-agentic-ai-deployment
  15. FDA press release – FDA expands AI capabilities and completes data platform consolidation (May 6, 2026): https://www.fda.gov/news-events/press-announcements/fda-expands-ai-capabilities-and-completes-data-platform-consolidation
  16. American Medical Association – AMA survey: Prior authorization reform pledge falls short with physicians (May 13, 2026): https://www.ama-assn.org/press-center/ama-press-releases/ama-survey-prior-authorization-reform-pledge-falls-short-physicians
  17. Pharmaceutical Executive – The prior authorization system fails patients. There’s a better way (May 14, 2026): https://pharmexec.com/view/the-prior-authorization-system-fails-patients-there-s-a-better-way
  18. IQVIA – From automation to oversight: building an AI-ready MLR function (Jul 2026): https://www.iqvia.com/blogs/2026/07/from-automation-to-oversight
  19. Jayatunga et al. – How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons. Drug Discovery Today 29(6), 2024: https://doi.org/10.1016/j.drudis.2024.104009
  20. Insilico Medicine – Insilico initiates Phase III clinical trial for rentosertib (Oct 7, 2026): https://insilico.com/news/xmjsn41091-insilico-initiates-phase-iii-clinical-tr
  21. Insilico Medicine – First patient dosed in GENESIS-IPF-3 Phase III trial of rentosertib (Sep 2026): https://insilico.com/news/an1009263-insilico-medicine-doses-first-patient-genesis-ipf-3
  22. Fierce Biotech – BenevolentAI’s AI-enabled drug flunks midphase eczema trial (Apr 2023): https://www.fiercebiotech.com/biotech/benevolentai-cruel-rd-ai-enabled-drug-flunks-midphase-eczema-trial-dent-deal-plans
  23. BenevolentAI press release – BenevolentAI unveils strategic plan to position the company for a new era in AI (May 25, 2023): https://www.businesswire.com/news/home/20230524005926/en/BenevolentAI-Unveils-Strategic-Plan-to-Position-the-Company-for-a-New-Era-in-AI
  24. Fierce Biotech – Verge Genomics drops sole clinical candidate in return to AI drug discovery roots: https://www.fiercebiotech.com/biotech/verge-genomics-drops-sole-clinical-candidate-return-ai-drug-discovery-roots
About the author
Shrish Dewangan

Shrish Dewangan

Principal

13+ years of pharmaceutical forecasting and analytics across US and global markets, with a strong focus on Rare Diseases alongside Cardiovascular, Renal and Diabetes. Previously led forecasting for Amgen’s US Rare Disease Business Unit, and has worked at Novo Nordisk and PharmaACE.

Prepared by PharmSight Research & Analytics for informational purposes only.