The Bottleneck in Cancer Research Isn't the Algorithm

Everyone's racing to build a better model. Almost nobody is asking whether the platform underneath it can actually keep up.

At Memorial Sloan Kettering Cancer Center (MSKCC), they're not debating benchmarks. They're running the #2 supercomputer in the United States and #4 globally, compressing research timelines that used to span years into months — and building the case that in cancer research, infrastructure isn't a support function. It's the whole game.

That's the argument Jessica Audette, Head of Enterprise Infrastructure Strategy and Transformation at MSKCC, makes to Dr. Serena Huang in the latest episode of Deep Geeks.

The algorithm is not the bottleneck. Infrastructure is.

Researchers have always had algorithms. What they didn't have (until recently) was a system fast enough to make those algorithms matter at speed.

When MSKCC set out to build IRIS, they didn't upgrade their hardware. They scrapped the old architecture and rebuilt it from the ground up around best-in-class solutions for AI and HPC workloads (yes, that includes NeuralMesh which closes the gap between compute and data). The result: 30x reduction in wall clock time. Years of discovery compressed into months.

Jess's analogy is the one that sticks: the researchers are the Ferraris. IRIS is the Autobahn. Without the road, even the best car just sits in the garage.

The unglamorous wins that change everything

The IRIS headline is dramatic. But some of MSKCC's most meaningful AI deployments are quieter.

Predictive scheduling that accounts for real transit time between New York City clinical sites — so a patient isn't expected to cross town in 10 minutes during rush hour. AI voice dictation so clinicians walk out of a room, say what they need to say, and move on — instead of writing charts at the dinner table like Jess's parents used to do.

And then there's Amelia. MSKCC's internal agentic AI on the IT help desk reduced call wait times from 42 minutes to one minute, with 35% cost savings. Brilliant technologists who were fielding password resets and printer tickets are now solving problems that actually need them. That's the version of "AI augments humans" that plays out in real life, not in a keynote.

Infrastructure investment is hard to justify. Until it isn't.

The harder problem at institutions like MSKCC isn't building the platform. It's proving it deserves continued investment.

Jess solved this the data way: she went back through every grant awarded to MSKCC and identified the ones where the high-performance computing platform was a deciding factor. The number she came back with was $184 million in research funding that came to MSKCC specifically because IRIS existed.

That's the translation layer most technology leaders skip — from platform to science, from science to funding, from funding to mission. Once you can show that number, the conversation about infrastructure investment changes completely.

What the industry gets wrong about innovation

Jess closes with a philosophy worth sitting with.

The Japanese art of kintsugi repairs broken pottery with gold. The cracks don't disappear — they become the most beautiful part of the piece. She sees AI the same way: the disruption, the fear, the breaking apart of how work has always been done — that's not the problem. That's the process. What gets built coming out the other side will be worth it.

The AI conversation usually focuses on what might break. This episode is about what's already being built — and how fast.

🎙️ Watch Deep Geeks Ep. 2 now “Cancer Doesn’t Care About Your Model” on YouTube or listen on Spotify or Apple Podcasts.

If this changes how you think about AI infrastructure, share it with someone who needs to hear it!