OpenAI says its AI models improved the heavily optimized flagship GPU designs of Advanced Micro Devices, better known as AMD, by up to 14% on power, performance and area, and that some of the model-generated changes are already in a chip design the company has sent for manufacturing

1

2

3

. OpenAI presents the result as the first concrete evidence that AI models can improve chips expert human teams have already designed and optimized

1

.

Source: AIM

Source: AIM

How the models did it

The models rewrote the SystemVerilog code that describes chip circuits. Software from Synopsys, the design-tools vendor, verified that each change kept the chip working and measured its effect on power, performance and area

1

3

. Greg Brockman, OpenAI's president, summed up the pace on X: "roughly an additional year of progress on @AMD's chip achieved in less than 5 months using our models"

3

. He called the results "unprecedented results at the frontier of chip design"

3

.

Source: The Next Web

Source: The Next Web

Beyond the GPU

The chipmaker independently verified single-digit percentage gains in key blocks of the upcoming next-generation flagship CPU, working from designs OpenAI's models produced

1

2

3

. OpenAI also applied the same approach to Jalapeño, its own first in-house chip. The chip, developed with Broadcom and unveiled in August, reached manufacturing nine months after design began

3

1

. Model-suggested changes cut two key components by about 13% and 10%, and those held up in finished silicon

1

3

.

The caution inside the announcement

Richard Ho, OpenAI's own chip chief, said last month he does not fully trust the model-generated design changes yet, so a human checks every one

1

. His caution does not stand alone: OpenAI itself acknowledged limitations in the methodology of its benchmark testing

3

. Its newest, GPT-6.2 Astra, scored highest both at improving existing designs and at building a processor from written requirements

1

3

.

The design work also sits inside a large commercial relationship. An agreement the two companies announced in October 2025 covers up to 6 gigawatts of Instinct GPU deployments, starting with 1 gigawatt in the second half of 2026. The chipmaker also handed OpenAI a warrant for as many as 160 million shares tied to milestones

2

3

.

Training a model for the design tools

OpenAI is training a joint model with Synopsys, which is refocusing on AI-driven chip design, to operate the software chip engineers already use

1

3

. Reuters reported in September that OpenAI had partnered with Synopsys on the model

3

. Both companies expect more gains across their GPU and CPU lines as the models keep optimizing against existing baselines

2

. Brockman has said the collaboration could shave weeks or months off development timelines, with the traditional verification tools continuing to check whatever the models write

3

.

© 2026 TheOutpost.AI All rights reserved