Skip to main content
Latest Update
2026-11-18 11:05 UTC

OpenAI finds 14% PPA gains on AMD GPU IP in 5 months

OpenAI finds 14% PPA gains on AMD GPU IP in 5 months

According to OpenAI via Greg Brockman’s post and the OpenAI blog, its models delivered up to 14% PPA improvement on AMD flagship GPU IP in under five months, with some results taped out and similar gains verified on OpenAI’s Jalapeño inference chip, while partnering with Synopsys to extend these chip design capabilities.

Source

Analysis

OpenAI announced on November 18 2026 that its models delivered roughly one additional year of progress on AMD chip designs in under five months with select improvements already included in a recently taped-out design.

According to OpenAI the models produced up to 14 percent PPA gains on flagship GPU IP building on years of prior AMD engineering work and delivered verified improvements on the Jalapeño custom inference chip using real silicon measurements. The same announcement details ongoing collaboration with Synopsys to integrate these model capabilities into established EDA tools. This development shows how AI accelerates chip design workflows by compressing multi-year optimization cycles into months thereby boosting engineer productivity and enabling faster iteration on hardware for inference and GPU applications across the broader AI ecosystem.

Greg Brockman

@gdb

President & Co-Founder of OpenAI