✉ The Friday AI Brief: the week's 5 best AI stories, tools & comparisons — in your inbox every Friday morning.

OpenAI logo

OpenAI’s Models Rewrote AMD’s Flagship GPU Circuits — and Some Are Already Headed to Manufacturing

OpenAI’s frontier models rewrote parts of AMD’s flagship GPU design and improved its power, performance and area (PPA) by up to 14% over baselines AMD’s own engineers had spent years optimizing — and AMD has already put some of the model-written changes into a chip design it recently sent for manufacturing. Why it matters: this is one of the first times an AI lab has published circuit-level results on a partner’s production silicon, not a benchmark toy.

November 18, 2026

What happened

In a technical report published November 18, “Advancing chip design with OpenAI models,” OpenAI described using its models to rewrite SystemVerilog RTL — the code that describes a chip’s circuits — on AMD’s flagship GPU intellectual property. The company says the rewrites delivered up to a 14% PPA improvement relative to heavily optimized baselines, work it equates to roughly a year of design progress achieved in less than five months. Each rewrite was checked with Synopsys design tools to confirm the circuit still behaved correctly and to measure the trade-offs.

AMD has included some of the changes in a design recently sent to manufacturing and written others into production design files for future versions, according to OpenAI. AMD also verified single-digit percentage PPA gains across key blocks of its upcoming next-generation flagship CPU using model-written designs.

The same approach, proven in OpenAI’s own silicon

OpenAI says its engineers have used models on chip design since early 2025, starting with GPT-4o on Jalapeño, the company’s first custom inference chip. It estimates model-proposed changes cut standard-cell area by about 13% in the chip’s vector compute unit and about 10% in its matrix unit. One detailed example: a multiplier redesign that used about 15% less cell area in synthesis while cutting latency from two clock cycles to one. OpenAI says those optimizations were proven in silicon during Jalapeño’s bring-up.

To track progress across model generations, OpenAI also ran 9,881 scored tests on two open-source RISC-V processor cores, PicoRV32 and Ibex. GPT-6.2 Astra posted the highest average scores at a comparable budget of about 112,000 tokens per run. OpenAI flags a caveat itself: models designing from written specifications could omit features the original cores include, which may flatter those scores.

Why it matters

Chip design is slow, expensive and short on senior engineers, so even single-digit efficiency gains on production designs are worth real money — and here the customer has already taped the changes into a manufacturing-bound design. The honest limits: every number is self-reported by OpenAI, the 14% is an “up to” figure for specific GPU IP rather than a whole-chip gain, and OpenAI’s post carries no statement from AMD. Still, the direction is clear, and OpenAI is productizing it: under a multi-year agreement with Synopsys announced September 30, the two are training GPT-Synopsys to run professional design tools the way expert engineers do. Everything AI, tested — and increasingly, everything AI-designed, too.

Frequently asked questions

What exactly did OpenAI’s models improve at AMD?

Power, performance and area (PPA) on AMD’s flagship GPU IP — by up to 14% versus heavily optimized baselines, per OpenAI — by rewriting the SystemVerilog RTL that describes the circuits.

Are the AI-written changes actually being manufactured?

OpenAI says AMD included some changes in a design recently sent for manufacturing, and saved others in production design files for future versions.

What is GPT-Synopsys?

A specialized model OpenAI and Synopsys are training, under a multi-year partnership announced September 30, to operate Synopsys’ electronic design automation tools for chip design and verification work.

Sources: OpenAI, Unite.AI, The Decoder via MLQ

Leave a Comment

Your email address will not be published. Required fields are marked *

Get the 5 best AI tools every week

Top AI news, tools, and prompts — one short email. Free, unsubscribe anytime.

Run a newsletter of your own? Monetize and grow it with SparkLoop →

Scroll to Top