The AI story sits on top of a very conventional calendar
The AI pitch in oncology is not what moves the next ninety days. The calendar is. Summit Therapeutics' Biologics License Application for ivonescimab, based on the HARMONi trial, has an FDA PDUFA goal action date of November 14, 2026.1 Moderna and Merck plan to present detailed results for their personalised mRNA cancer vaccine, intismeran, at ESMO on October 24, and to engage with regulators about potential filings.2 On October 2, Summit announced a trial collaboration with Daiichi Sankyo and AstraZeneca to test ivonescimab with Datroway across multiple solid tumour settings, intending to include breast and lung cancer.3
For readers who follow AI, the useful question is how much of the next-generation oncology pipeline is AI-derived and how much only shares the same moment. On the evidence we hold, the answer is that the dossier does not tie any of these three programmes to AI design. The AI material comes from a separate source: an AstraZeneca executive describing how the company wants to work. We keep the two threads apart below.
What AstraZeneca says AI will do for biologics
The AI argument comes from Puja Sapra in a piece published by MIT Technology Review, "How AI helps scientists design the next generation of medicines." It is the company's own account of its method, so it should be read as such.4
The central claim is about data. 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."4 In other words, the claimed edge is not a model but the proprietary records of molecular structures, binding measurements, safety profiles and manufacturing outcomes behind it.
The second claim is about automation. Sapra describes a 'lab of the future': "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."4 She adds that this "will generate AI-ready data at a scale that traditional workflows cannot match," and that robotic sample handling, automated quality checks and integrated data pipelines could "help accelerate early drug development timelines significantly."4 Note the hedge: 'have the potential to'. No timeline reduction is quantified in what we hold.
The third claim ties back to the combination trend above. For multi-specific biologics, Sapra 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."4 That is the same design problem a bispecific like ivonescimab poses. The article does not say AI was used on it.
The hard part, in the speaker's own words
The most credible passages are the admissions. Sapra writes: "One of the hardest problems in de novo design is predicting whether a computationally generated molecule will be safe in the human body."4 Her proposed answer is virtual clinical trials using advanced cell systems and micro-scale organ models, which "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."4 The phrase 'missing piece' concedes that the loop between design and clinic is not yet closed. She also says that as systems become more autonomous, "human oversight remains at the heart of this approach—ensuring explainable and ethical AI for the benefit of patients."4
How much weight the AI source can bear
Via News measures the reliability of the sources it ingests. For this MIT Technology Review item, 0% of 11 checked claims held up.4 That does not mean Sapra's statements are false. The quotes above are verbatim, and we attribute them to her. It means the claims we extracted from the piece could not be confirmed against a source of truth. They are intentions and positioning, not results. The article is also undated in our records, although its page header shows 2026-07-23. A reader should treat 'AI cuts discovery timelines' as a stated aim, not a demonstrated outcome.
The money side: Merck's verified numbers
Merck is the one company here with SEC-verified financials. They show a large and uneven balance sheet, not a clean 'AI spend' signal, and nothing in the filings data attributes any figure to AI or vaccines.
Capital expenditure was $3.863 billion in fiscal 2023, $3.372 billion in 2024 and $4.112 billion in 2025.5 The 2025 figure is about 22% higher than 2024. Cash swung much more. It was $13.242 billion at the end of 2024 and $14.565 billion at the end of 2025.5 It stood at $18.169 billion in Q3 2025, then $5.327 billion in Q1 2026 and $6.849 billion in Q2 2026.5 From the Q3 2025 peak to Q2 2026, that is a fall of roughly 62%. The filings data we hold does not explain the drop, so we do not. Cost of revenue was $4.395 billion in Q2 2026 against $3.557 billion in Q2 2025.5
Capital is uneven across the sector
The smaller companies show the other end. ADC Therapeutics cut jobs, saying the reduction is "projected to deliver $10 million in annualised cost savings," and that it "maintains a cash runway into 2028, supporting ongoing regulatory and clinical plans for ZYNLONTA combinations."6 Novocure's chief executive Frank Leonard called its second quarter 2026 "our strongest quarter to date, with record net revenues and active patients on therapy."7 The company reported net revenues of $184 million, up 16% year over year.7
Reliability matters here too. The press-release feeds behind these items scored 56% on checked claims for the Novocure feed and 45% for the weekly roundup.6,7 Company statements are the company's own framing.
Combinations are still a clinical story
Elsewhere in the field, the European regulator's committee (CHMP) recommended Gilead's Trodelvy plus Keytruda for first-line PD-(L)1-positive metastatic triple-negative breast cancer.8 Telix dosed the first patient in its Phase 3 LUTEON study of TLX250-Tx in renal cancer.9 And on October 8, the EXCALIBER-RRMM Phase 3 reported topline results: median progression-free survival of 42 months for ZDd against 20 months for DVd (hazard ratio 0.49, p<0.000001), a 51% reduction in the risk of progression or death, with safety consistent with prior findings.10
What to watch
- October 24: intismeran's detailed ESMO data, and whether Merck and Moderna announce regulatory plans.2
- November 14: the FDA's PDUFA decision on ivonescimab.1
- Any measured AI result: a stated reduction in discovery time, or a named AI-designed molecule in the clinic. The AstraZeneca piece offers neither, so a concrete number would be the first real evidence.
- Merck's next filing: whether cash rebuilds from $6.849 billion.5