The AI claim, and how much weight it can bear
The most concrete description of AI inside big-pharma oncology discovery in our dossier comes from Puja Sapra in an MIT Technology Review piece on how AI helps scientists design new medicines. She describes a 'lab of the future' and compares it to autonomous driving: "Where a self-driving car uses sensors and models to navigate its environment, this system uses AI to make predictions, robotic systems to execute experiments, and instruments to generate data."1
Our trust check matters here. Via News measured the reliability of that source, and none of the 11 claims we checked from it held up.1 That does not mean the quotes are misattributed. It means none of the underlying assertions have been independently confirmed, and the piece is undated in our records.1 Treat everything below from this source as a company's own account of its ambitions, not as evidence of results.
Data as the moat
The argument for why AstraZeneca expects to win is about data, not algorithms. Sapra says: "Data is our differentiator. We've built an intentionally diverse portfolio across multiple disease areas and drug types. All of that data empowers us to fine-tune frontier AI models with richer, more representative training sets."1 The dossier ties this to proprietary datasets covering molecular structures, binding measurements, safety profiles and manufacturing outcomes.1
She also says automation will "generate AI-ready data at a scale that traditional workflows cannot match," with robotic sample handling and integrated data pipelines that could "help accelerate early drug development timelines significantly."1 For practitioners, the point is that the bottleneck being targeted is experimental throughput. A model is only as good as the lab loop that feeds it.
The hard part: safety, and the multi-specific problem
The source is candid about what AI has not solved. Sapra states: "One of the hardest problems in de novo design is predicting whether a computationally generated molecule will be safe in the human body."1 Her proposed answer is virtual clinical trials using advanced cell systems and micro-scale organ models. She says they "have the potential to generate enhanced biological signals without traditional testing bottlenecks, and they're a critical missing piece in closing the loop between AI-generated designs and clinical-ready candidates."1 The wording is 'potential', and the claim is forward-looking.
For multi-specific biologics, she says such models "could help identify which two or three targets to prioritize based on the underlying biology, then optimize across multiple parameters to balance a molecule's potency, stability, manufacturability, and safety."1 On oversight, she adds: "With more autonomous systems, human oversight remains at the heart of this approach—ensuring explainable and ethical AI for the benefit of patients."1
Taken together, these quotes describe a design-and-test loop that is not yet closed. The step from generated molecule to human safety is named as unresolved by the same source that promotes the approach.
Where the pipeline actually is: dates that can test the thesis
The clinical calendar offers real checkpoints, though none of them is a test of AI design. Moderna and Merck plan to present detailed results for their personalized mRNA vaccine, intismeran, at ESMO on Oct. 24 and to engage with regulators on potential filings.2 Our event record from Aug. 19 says the vaccine stopped melanoma from returning in a historic clinical trial.2 The detailed data will show how large and durable that effect is.
Summit Therapeutics has a Biologics License Application for ivonescimab, based on its HARMONi trial results, with an FDA action date of Nov. 14, 2026.3 On Oct. 2 Summit announced a trial collaboration with Daiichi Sankyo and AstraZeneca to test ivonescimab with Datroway across multiple solid tumor settings, intended to include breast and lung cancer.3 That combines an antibody-drug conjugate with a bispecific antibody. It is a case where the multi-target logic Sapra describes is already being tested clinically, though the dossier does not say AI was used to design either drug.
Elsewhere in breast cancer, the European Medicines Agency's CHMP recommended Gilead's Trodelvy plus Keytruda in PD-(L)1-positive first-line metastatic triple-negative breast cancer.4 Evandro de Azambuja said patients "need effective treatment options as early as possible in the course of their disease, particularly when their tumors express PD-L1."4 This is a combination-therapy trend that does not depend on AI.
The money side: steady spending, not a visible AI step-change
Merck's SEC-filed figures are canonical and fidelity-checked. Full-year capital expenditure was $3.863 billion in 2023, $3.372 billion in 2024 and $4.112 billion in 2025.5 That 2025 figure is roughly 22% above 2024, by our arithmetic from those two filings.5 First-quarter spending ran $861 million in 2024, $1.328 billion in 2025 and $991 million in 2026.5
Cash swung widely. It was $6.841 billion at the end of 2023, $13.242 billion at the end of 2024 and $14.565 billion at the end of 2025.5 Quarter-end cash reached $18.169 billion in Q3 2025, then stood at $5.327 billion in Q1 2026 and $6.849 billion in Q2 2026.5 For scale, the Q3 2025 balance is more than double the cash held a year earlier at the end of Q2 2025 ($8.007 billion).5
Cost of revenue rose from $15.193 billion in 2024 to $16.382 billion in 2025.5 In Q2 2026 it was $4.395 billion, against $3.557 billion in Q2 2025, about 24% higher.5 The honest reading is limited. These filings show a company spending steadily and holding large cash balances. They do not tell us how much, if any, of the spending is AI-related, and the dossier does not say. We will not infer it.
Smaller companies: commercial wins and cost pressure
Novocure reported second-quarter net revenues of $184 million, up 16% year over year, driven by 18% global active-patient growth.6 It received a CE Mark for Optune Pax in locally advanced pancreatic cancer, with more than 280 active patients on it as of June 30, 2026.6 Frank Leonard called it "our strongest quarter to date, with record net revenues and active patients on therapy."6
The picture is less bright at ADC Therapeutics. It announced a workforce reduction "projected to deliver $10 million in annualised cost savings" and says it "maintains a cash runway into 2028, supporting ongoing regulatory and clinical plans for ZYNLONTA combinations."7 Telix dosed the first patient in a Phase 3 trial of TLX250-Tx in relapsed or recurrent clear cell renal cell carcinoma, the first radiopharmaceutical therapy to enter Phase 3 for that cancer.8 Both are sound developments, but neither involves AI.
Source reliability
The wire-service releases carry their own caveat. Via News's measured reliability for the newswire source behind the Gilead and Novocure items is 56% of 4,956 checked claims holding up.4,6 For the Telix source it is 31% of 2,927,8 and for the weekly sector roundup carrying the ADC Therapeutics item it is 45% of 520.7 These are company announcements, so the quotes are accurate to the speaker, but forward-looking statements inside them should be weighed accordingly.
What to watch
- Oct. 24: the detailed intismeran data at ESMO, and any word on regulatory filings.2
- Nov. 14: the FDA action on Summit's ivonescimab BLA.3
- Evidence of AI results: the 'lab of the future' framework needs disclosed outcomes, such as molecules advancing with fewer failures. The dossier contains none yet, and the safety-prediction gap Sapra names is the one to track.1
- Merck's next filings: whether capex, last reported at $4.112 billion for 2025, keeps rising.5