AI doesn't need giant supercomputers after all
I attended the 2026 Salishan Conference on High Speed Computing last month, and it was a week well spent in coastal Oregon hearing what many of the world's experts in scientific computing are worried about. Although I took copious notes and many of the presentations are online now , I'm not really sure how closed-door the week is meant to be. As such, I don't feel quite comfortable sharing a distillation of my observations (beyond what I posted to Bluesky that week) without someone telling me I can. However, I do feel comfortable sharing a presentation that I gave. I wasn't asked to give this talk; rather, I was allowed to present it as part of Salishan's fun "Random Access" session, a round of lightning talks that are both anonymously submitted and voted on in the first two days of the conference I submitted a talk titled "AI doesn't need giant supercomputers after all" based on a half-written blog post I've been struggling to finish,...
