Cancer medicine has had a busy autumn. On October 9 the US Food and Drug Administration approved Roche's Tecentriq, in combination with two chemotherapy drugs, as a treatment after surgery for stage III colon cancer whose tumours have a specific genetic fault known as dMMR. It is the twelfth US use approved for the drug.1 The approval rests on a Phase III trial called ATOMIC.1 It is the clearest verified regulatory win in this set of events.
Behind the approvals sit two other stories: how much money it takes to bring a cancer drug to the finish line, and how far artificial intelligence has actually got in designing one. Both are worth reading with care. The most hopeful statements here are forecasts, and one of the sources has a poor measured accuracy record.
A run of regulatory milestones
Europe is moving too. On October 16 the European Medicines Agency's drug-review committee (the CHMP) gave a positive opinion on a new dosing schedule for Libtayo across all its approved uses. The schedule is 700 mg as a 30-minute infusion every six weeks, after an initial 12 weeks of 350 mg every three weeks.2 A positive opinion is a recommendation. It is not yet a marketing authorisation.
In July the same committee recommended Gilead's Trodelvy together with Merck's Keytruda for first-line treatment of one type of metastatic triple-negative breast cancer.3 Evandro de Azambuja said patients with this cancer need effective options “as early as possible in the course of their disease.”3 That comment was part of the company's announcement. Treat it as a clinician's view carried by the drugmaker, not as independent assessment.
Two more dates are close. Moderna and Merck plan to present detailed results for their personalised cancer vaccine candidate, intismeran, at the ESMO cancer conference on October 24. They also plan to engage with regulators about potential filings.4 The FDA has set November 14, 2026 as its decision date for Summit Therapeutics' application for ivonescimab, which is based on the HARMONi trial.5 The dossier does not say what the trial found, so we do not characterise it.
What Summit's accounts show
Summit is the clearest case in this set where we can check the money against SEC filings. The numbers below come from Via News's verified data, which is checked against the filed accounts.
The company reported no revenue in 2023 or 2024.6 Its net loss was about $615 million in 2023 and about $221 million in 2024. In 2025 it was about $1.08 billion.6 That is roughly five times the 2024 loss. The dossier does not explain what drove the swings, so we do not attribute them.
Cash spent running the business tells a steadier story. Operating cash flow, meaning cash actually going out of the business before any fundraising, was minus $76.8 million in 2023, minus $142.1 million in 2024 and minus $322.9 million in 2025.6 The outflow more than doubled in 2024 and again in 2025. Spending on equipment was tiny by comparison, at $657,000 in 2025.6 The money is going to running the company and its trials, not to building things.
Cash on hand moved in large steps:
- $211.5 million at the end of Q1 2025.6
- $297.9 million at Q2 2025.6
- $238.6 million at Q3 2025.6
- $225.3 million at year-end 2025.6
- $106.5 million at Q1 2026.6
- $419.4 million at Q2 2026.6
The filings in our data show the balance recovering sharply after Q1 2026. They do not say where the money came from, and we will not guess.
Here is our own arithmetic, not a forecast. Year-end 2025 cash of $225.3 million was smaller than the $322.9 million the company spent on operations during 2025.6 The $419.4 million at Q2 2026 equals roughly 1.3 years at that 2025 rate.6 A company that spends more than it holds, with no sales yet, depends on the decision due November 14 and on its ability to raise money.
Deals and consolidation
The topic summary for this piece also refers to McKesson and the private-equity firm CD&R acquiring Option Care Health, which provides infusion services. It also refers to a Summit, AstraZeneca and Daiichi Sankyo collaboration testing ivonescimab with the drug Datroway in first-line triple-negative breast cancer. The dossier holds no source document for either deal and no financial terms, so we can only report that they were flagged. We cannot verify them or describe their size.
Elsewhere in the sector the commercial picture is mixed. Novocure reported second-quarter 2026 revenue of $184 million, up 16% from a year earlier, with 18% growth in active patients worldwide.7 Chief executive Frank Leonard called it “our strongest quarter to date, with record net revenues and active patients on therapy.”7 That is the company describing its own results.
Talent is also moving. Everest Medicines hired two former Moderna oncology scientists in one day, October 9: Brandon Coder, formerly of cancer vaccines, and Lakshmi Srinivasan, formerly head of oncology translational medicine.8
Real-world evidence on an older drug
Not every useful result is a new launch. Kyowa Kirin announced final results of the German MINT study of mogamulizumab in 109 patients with two rare blood-related skin cancers, mycosis fungoides and Sézary syndrome, across 18 centres.9 The median time before patients needed another treatment was 14.9 months. The best overall response rate was 62.4%, and median progression-free survival was 20.2 months.9 These are real-world figures from routine care, not a controlled trial, and they come from the manufacturer's announcement.
The AI claim: a vision, not yet a result
AstraZeneca's Puja Sapra, in an article in MIT Technology Review, describes where AI could take drug design. Her central idea is 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.”10 She says AI models could help pick “which two or three targets to prioritize” and balance a molecule's “potency, stability, manufacturability, and safety.”10
She is also candid about the limits: “One of the hardest problems in de novo design is predicting whether a computationally generated molecule will be safe in the human body.”10 She argues that AstraZeneca's data is its edge: “Data is our differentiator.”10 And she says “human oversight remains at the heart of this approach.”10
Read these as forward-looking statements from a company executive. The article names no drug that AI designed and no trial result. The source also matters: our checks found that none of the 11 claims we tested from this MIT Technology Review article held up.10 That is a small sample, and the zero says nothing about Ms Sapra's honesty. It does mean this piece should not be treated as evidence that the technology works.
How much to trust the sources
Most of the press releases here come through a wire feed whose claims have held up 56% of the time across 4,960 checks.3,7,9 That means roughly four in ten checked claims did not hold up. We have therefore leaned on the SEC-filed numbers for Summit and described company statements as the companies' own.
What to watch
- October 24: detailed intismeran data at ESMO, and any signal about regulatory filings.4
- November 14: the FDA decision date for ivonescimab. Summit's cash ($419.4 million at Q2 2026) against an operating outflow of $322.9 million in 2025 frames how much rides on it.5,6
- Libtayo: whether the European Commission follows the committee's recommendation.2
- AI: the first named medicine where AstraZeneca can show an AI-designed molecule passing a safety test. Until then, the AI story remains a forecast.
