The Opening
This week, the business of inventing medicines went looking for more horsepower, and found it.
One of the world's biggest drugmakers switched on its second AI supercomputer. A Nobel laureate's lab unveiled gene editors that evolution never got around to building. Regulators opened the door to letting algorithms sit inside live clinical trials. And a Broad Institute team taught an AI to hand laboratory evolution a running start.
It is a lot of firepower pointed at the same stubborn fact: turning a promising molecule into a medicine still takes years, still runs through human trials, and still, so far, has not produced a single AI-designed drug that a regulator has approved.
So this Sunday we are asking a quieter question underneath all the compute. Not how fast can we design a drug, but what actually stands between a clever molecule and the patient who needs it. The answer, it turns out, is rarely the chips.
Let's get into it.
- Troy, Ray, and Ibrahim
The AI factories are getting bigger. The clinic is still the bottleneck.
On July 20, Bristol Myers Squibb and NVIDIA announced that BMS is switching on its second AI supercomputer, a cluster built on eight of NVIDIA's new DGX Vera Rubin systems that the companies call the most powerful in the life sciences. The pitch: up to ten times more performance per megawatt than the machine it replaces, and access opened, in the words of one BMS vice president, to "literally every scientist" rather than a select few.
It is a genuinely big machine. It is also, as STAT drily noted, the third time in nine months that a pharma company has announced it is assembling the largest AI supercomputer in the industry.
Why it matters: Compute is now the easy part. The jobs BMS is pointing this cluster at, target identification, molecule design, and digital twins, really can save scientists weeks of work. Faster, cheaper starting points are real, and they compound.
The catch: None of it touches the part that actually takes the years. More than 170 AI-originated drugs are now in clinical development, and as of this spring not one discovered end to end by AI has won FDA approval. The furthest along, Insilico's rentosertib, only entered Phase III this month. Trials still need patients, time, and a bit of luck.
Bottom line: The chips were never the bottleneck. The clinic is. A supercomputer can hand you a thousand candidate molecules by Friday, but it cannot enroll a trial, and it cannot tell you which one helps a person feel better.
Pharma keeps building bigger AI supercomputers to design drugs. What will actually decide whether they reach patients?
We're collecting stories. This week, one question.
All that discovery firepower eventually lands in a clinic, a pharmacy, or a research ward, where someone actually has to run it. This week, that is where we are pointing.
Tell us about a drug trial, a new therapy, or a research tool that looked brilliant on paper and then met real patients. Maybe you enrolled people in a study and watched the protocol collide with their actual lives. Maybe you dispensed a breakthrough that the data loved and the patient could not tolerate. Maybe you are the coordinator, pharmacist, or lab tech who saw the gap between the press release and the bedside.
We want the real version, not the conference-panel version. Two paragraphs is plenty.
This is open to every corner of medicine: EMS, nursing, pharmacy, techs, registration, environmental services, and administration, all of it. You choose how you're named, whether that's full name, role only, or fully anonymous. We protect patients in every story we run. That is not negotiable and it never will be.
If we run your story, we'll send you a Consult mug as a thank you.
Dr. David Fajgenbaum, Co-Founder and President, Every Cure; physician-scientist, University of Pennsylvania
Fajgenbaum's interest in unused medicine is personal. As a medical student he nearly died from idiopathic multicentric Castleman disease, an immune disorder with few good options, and eventually found that an already-approved drug could hold his own illness in remission. He is still here, years later, because of a medicine that already existed. In 2022 he co-founded Every Cure, a nonprofit built on a simple, unglamorous idea: there are thousands of approved drugs and thousands of diseases, and almost nobody has systematically checked which forgotten pairings might work.
Where things stand:
Every Cure's MATRIX platform uses AI to score existing FDA-approved drugs against thousands of diseases, ranking the repurposing matches most worth testing
Selected in February 2026 for up to $76 million from ARPA-H, building on an earlier $48.3 million award, to move AI-flagged candidates toward patients
10 active repurposing programs today, from breast cancer to Bachmann-Bupp syndrome, a condition affecting roughly 20 people worldwide
The new phase aims to fund preclinical work on at least 20 candidates and clinical trials for 10
Illustration: The Consult
Why they matter: While the industry spends billions teaching supercomputers to invent new molecules, Fajgenbaum is chasing the drugs we already have and never fully used. It is the least glamorous corner of AI drug discovery, no novel chemistry and no room full of GPUs, and possibly one of the most humane: matches that could reach patients in years, not decades, because the safety data already exists.
The FDA wants AI inside the trial
The FDA is piloting "real-time" clinical trials that stream data straight to reviewers as it is collected, with proof-of-concept studies already running in AstraZeneca's mantle cell lymphoma trial and Amgen's small cell lung cancer trial. A parallel program invites AI to help with recruitment, dosing, and safety monitoring, and the agency's AI chief says it could cut 20 to 40% off trial timelines. Final selections land in August.
An AI drug shop bolts on a factory
Insilico Medicine, whose AI platform has nominated 31 preclinical drug candidates since 2021 and cleared 13 into human testing, signed a July 15 alliance with contract manufacturer Bora Pharmaceuticals. The logic: designing molecules fast means little if you cannot make them at scale. Insilico says its AI shrinks early discovery from the usual 2.5 to 4 years down to 12 to 18 months.
AI designed a gene editor evolution never made
In Science, a team including Nobel laureate Jennifer Doudna used AI protein design to engineer synthetic versions of TnpB, a compact cousin of the CRISPR enzymes, with sequences far from anything found in nature. The best variants edited human and plant cells as well as or better than the natural editor. It is a hint that the gene-editing toolkit no longer has to wait on evolution.
AI gives lab evolution a head start
A Broad Institute team led by David Liu reported in Nature that starting protein evolution from AI-stabilized designs, rather than from natural proteins, produced far better results. One evolved enzyme cut a neurodegeneration-linked protein 79 times more effectively than versions grown from the natural starting point. Better building blocks, better medicines.
The movie always ends at the breakthrough
In 2010, Harrison Ford and Brendan Fraser starred in Extraordinary Measures, a film based on the true story of John Crowley, a father who quit his job and started a biotech company to chase a treatment for his two children's Pompe disease. It is an earnest movie, and like almost every drug-discovery story we tell ourselves, it ends more or less at the moment of triumph: the therapy works, the kids are saved, roll credits.
That is the cultural template. The lone genius, the eureka, the breakthrough. This week's headlines run on the same fuel, just industrialized: bigger supercomputers, faster molecules, gene editors conjured out of thin air. The promise is always compression, the ten-year slog squeezed into eighteen months.
But the movies end where the hard part begins. Crowley's real story did not stop at the discovery. It ran on for years through trials, regulators, manufacturing, and the agonizing wait of families who needed the drug before the process was ready for them. AI can genuinely shorten the first act. It has barely touched the second.
Which is what makes someone like David Fajgenbaum worth watching. He is not chasing a cinematic new molecule. He is combing through the drugs we already have, looking for the match that could reach a dying patient this year instead of next decade, because the waiting is the part that actually costs lives.
So here is the question the trailers never ask. When the models can sketch a cure in an afternoon, will we build the patience, and the systems, to get it to the person still waiting at the end of the hall?
Until next Sunday,
For the people keeping medicine human.
