EXPLORINEER
en

JP — ITmedia

OpenAI's AI Model Improves AMD GPU Circuit Design by Up to 14%, Also Yields Results for Its In-House "Jalapeno" Chip

AI2 min read

OpenAI says its AI models helped improve parts of a next-generation AMD GPU chip design by up to 14%. The work also helped design OpenAI's own inference chip, called Jalapeno.

What happened

On November 18, 2026 (US time), OpenAI announced that it used its AI models to improve chip designs. The models rewrote computer code that describes how a chip's circuits work, while keeping the circuits working correctly. The result was up to 14% better PPA in some circuits of AMD's next flagship GPU. OpenAI says this gain took less than five months and matched about one year of normal progress on an already well-tuned design.

Key numbers and facts

  • Up to 14% better PPA (power, performance, and area) in parts of AMD's next flagship GPU.
  • The 14% applies to specific circuits. It does not mean the whole GPU is 14% faster.
  • AMD used some of the ideas in a chip design now going to manufacturing, and others in future design data.
  • For OpenAI's Jalapeno chip, the model's changes cut the standard-cell area of the vector unit by about 13% and the matrix unit by about 10%. These are OpenAI's estimates.

Background

PPA is a way to score chip design quality. It looks at how much power a chip uses, how fast it works, and how much space its circuits take up. Chip designers use tools from Synopsys, a large company that makes software for designing electronics, to check that changed circuits still work and to balance power, speed, and size. OpenAI also tested its models on two open-source RISC-V processor cores. In that test, GPT-6.2 Astra got the highest average score on both cores when optimizing existing designs.

Why it matters

Chips are a key part of AI, and designing them takes many skilled engineers and a lot of time. If AI models can find real improvements faster, chips could get better sooner. OpenAI also noted a caveat: when a model designs a core from a written spec, it may drop some features from the original, so part of the score difference may come from that. Even so, the results show AI can help with real chip work.

What to watch

  • OpenAI plans to train a new model, called GPT-Synopsys, that can use design tools the way a chip engineer does.
  • Whether the gains appear in AMD's products when they reach customers.
  • How well the Jalapeno chip, made with Broadcom and announced in June, performs once it is in use.
  • How OpenAI and AMD's partnership, announced in October 2025 to build up to 6 GW of AI infrastructure, develops.

Glossary

PPA
A score for chip design quality, based on power use, performance, and circuit area.
GPU
A chip that handles many calculations at once, often used for graphics and AI.
Synopsys
A large company that makes software tools for designing electronic chips.
Jalapeno
OpenAI's first own AI chip, built for running AI models (inference), made with Broadcom.
RISC-V
A free, open design standard for computer processor chips.

Ad

Add 150 words. Learn 10 a day.

Dicter word books send only the words you learn to your review notes.

AI전자신문

Lablup Posts 48% Year-on-Year Revenue Growth for Nine Months, First Earnings Report Since Listing

Lablup reported cumulative consolidated revenue of 6.99 billion won and operating profit of 900 million won for the first three quarters, up 48.2% and 36.5% year on year, driven by corporate demand for AI infrastructure. It posted a cumulative net loss of 2.4 billion won due to a 3.49 billion won derivative liability valuation loss tied to the conversion of RCPS shares, but returned to a net profit of 360 million won in the third quarter as that accounting effect expired.

AIZDNet Korea

Maum AI Signs Physical AI Industry-Academia MOU with Hanyang University ERICA and Doosan Robotics

Maum AI signed a three-way MOU with Hanyang University ERICA and Doosan Robotics at the "AI Symposia" held on the 18th at Hanyang University ERICA to build cooperation on physical AI talent development and industry-academia collaboration. The partnership is based on the AX Graduate School project, a government-backed AI talent program run by the Ministry of Science and ICT, with cooperation planned in joint curricula, industry projects, internships, joint R&D and demonstrations, and government R&D planning. Maum AI will contribute its generative and multimodal AI and AI platform capabilities to the effort.

AIheise online

Google Genie 4: Interactive AI Worlds Now Available to Subscribers

Google is rolling out its Genie 4 AI world model to Google AI Ultra subscribers via the Project Genie web app, with access also available in Germany. Users of the cheaper Google AI Pro plan can join a waitlist, and daily usage limits apply. Genie 4 generates interactive 3D environments from text or images, and the model currently supports only about one minute of continuous interaction.

AIheise online

Gemini 4 Neon Challenges Google's Flagship Model

Google DeepMind has released Gemini 4 Neon, the fourth model in the Gemini 4 family within seven weeks, designed for efficiency while matching and slightly outperforming flagship model Carbon on some benchmarks. Neon is slated to become cheaper per token than Carbon from December, when Carbon's introductory prices expire. Google positions it for complex programming, analytical reasoning and enterprise use, filling the gap between Carbon and the everyday model Hydrogen.