LNA graphic
SAN FRANCISCO, November 18, 2026 — OpenAI says its artificial intelligence models have improved the design of AMD’s upcoming flagship graphics processor by up to 14 per cent on a standard industry measure, and that some of those changes have already been included in a design sent for manufacturing.
In a blog post published on November 18, the company said it had partnered with AMD on improvements in PPA — power, performance and area, the three numbers chip engineers use to judge whether a design is better, faster and cheaper to make. The models, it said, achieved up to a 14 per cent improvement relative to heavily optimised baselines on the flagship GPU intellectual property, by rewriting the SystemVerilog RTL code that describes the chip’s circuits.
RTL, or register-transfer level, is the detailed blueprint of how a chip moves data and carries out calculations. OpenAI said its model identified further improvements by rewriting that code, “at the level of an expert microarchitect”, and used Synopsys electronic design automation tools to check that each rewrite still behaved as intended and to measure the trade-offs in power, performance and area. The company called the result roughly a year of progress achieved in less than five months.
AMD, OpenAI said, has already included some of the proposed changes in a recent design it sent for manufacturing, and has added others to its production design files for use in future versions. The company also said AMD had verified single-digit percentage PPA gains across key blocks of its upcoming next-generation flagship central processor, using designs written by the model. It expects further gains on top of both the GPU and CPU relative to the most optimised baseline.
The post also described earlier work on OpenAI’s own custom inference chip, Jalapeño, and tests in which newer models, including GPT-6.2 Astra, scored more highly when asked to improve or redesign open-source RISC-V processor cores. Those experiments are laboratory scores, not shipping products, and should not be confused with the AMD claims.
The figures come from OpenAI itself and have not been independently audited in the reporting available today, so they should be treated as the company’s account of a private collaboration. Even so, the idea that a model can rewrite production chip circuitry, pass industry-standard checks and go into a tape-out is a notable claim for an industry that has long been short of specialist design engineers.
What to watch is whether the 14 per cent figure shows up in AMD products that actually reach the market, and how widely Synopsys and other toolmakers build similar assistants into everyday chip design. For Malaysia, which is deepening its role in semiconductor packaging, testing and, increasingly, design, the practical question is how quickly local engineers will be expected to work alongside these tools — and whether the gains accrue to the same global firms that already dominate the most advanced chips.
If the collaboration holds up under independent scrutiny, it would be one of the clearer examples yet of generative models moving from writing software to helping design the silicon that runs them. If it does not, it will join a growing list of ambitious AI-for-engineering claims that looked better in a blog post than on the factory floor.
Source: OpenAI, “Advancing chip design with OpenAI models”, November 18, 2026.

