OpenAI's Jalapeño chip wasn't AI-designed: 100+ ex-Google TPU engineers and Broadcom were
JFPuget · x · 2026-09-17
Pushing back on the narrative that OpenAI's Jalapeño chip was designed autonomously by AI, engineer jmartinprin lays out the real story: OpenAI hired 100+ engineers from Google's TPU team, co-designed the chip with Broadcom, benefits from faster HBM4 vs Blackwell's HBM3E, and made aggressive architectural tradeoffs (no complex scheduler) that will hurt it in scenarios like model disaggregation. The models mainly accelerated development rather than inventing the architecture. Context: an investor liquidated his Nvidia position arguing self-improving AI will erode Nvidia's moat.
More from Companies & People
- Company naming is the cheapest asymmetric bet founders can make, argues viral thread — alexmacgregor__ · 2026-09-17
- "Follow the Money": A Timeline Accuses Anthropic of Funding-IPO-Safety Club Rules — alex_verem · 2026-09-17
- Perplexity opens its first Berlin AI research roles to mine its search index — bo_wangbo · 2026-09-17
- NYT covers recursive self-improvement; Schmidhuber's ex-PhD student builds RSI startup Inherent — SchmidhuberAI · 2026-09-17
- Cloudflare's innovator's dilemma: protecting the human web may cost it the agentic one — EdenEmarco177 · 2026-09-17
- Stratechery Interviews Joanna Stern on the iPhone Duo Experiment and AI for Normal People — Stratechery · 2026-09-17