A GitHub repo named openTPU is testing a claim too big to just nod along to: that AI can now design its own inference hardware.
The project surfaced on Hacker News on October 6, framed around the idea that AI can now build its own inference hardware, and linking out to an open-source GitHub repository called openTPU. The name nods to Google's Tensor Processing Units, the custom chips built to run AI inference workloads faster and cheaper than general-purpose processors. Beyond the repo itself and that framing, there is not yet much public detail to go on. Within a day, the discussion had pulled in 78 points and 41 comments, a solid showing for a chip-design story on a forum where skepticism is usually the default reaction.
Chip design is one of the hardest jobs to hand off to AI: it demands deep domain expertise, long feedback loops, and little tolerance for subtle errors. If a model or agent genuinely contributed meaningful design work here, that is a real data point in the slow erosion of hardware engineering as an AI-proof job. If it did not, this is another case of a bold claim outrunning the evidence behind it.
Either way, the code is public, so unlike most grand AI claims, this one can actually be checked instead of just taken on faith.