OpenAI deploys Jalapeño ASICs on AMD EPYC Turin hosts
OpenAI chose AMD EPYC Turin over Nvidia Vera for the rack-scale setup, citing lower risk and maturity.
OpenAI says its Jalapeño ASIC was designed with heavy AI assistance and reached tapeout in nine months, which it presents as a new speed baseline for chip development. Richard Ho, OpenAI’s head of hardware, said the company used its own models and Codex throughout the process, while still relying on standard EDA sign-off flows before tapeout. Ho also said the team did not replace engineers, but made a smaller group far more productive, and noted that industry interest in the approach is already growing. Separately, OpenAI is deploying Jalapeño on AMD EPYC Turin hosts with 1.5TB of memory, choosing Turin over Nvidia’s Vera for maturity and lower risk in the rack-scale system.
Why it matters
This shows OpenAI is pairing its in-house chip effort with a more conservative host platform, rather than waiting for Nvidia’s Vera to mature. For teams following the company’s hardware push, the shift points to a deployment strategy built around reducing risk in the system around Jalapeño, not just speeding up chip design itself.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.