🔍 Read the full analysis: OpenAI’s AI Design Gains For AMD: The Results And Their Boundaries on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
OpenAI said on Nov. 18 that its models rewrote parts of AMD chip designs, reporting up to 14% gains in power, performance and area in a flagship GPU design. The reported best case is not an average, and neither the relevant chip blocks nor an independent audit have been disclosed.
OpenAI said Nov. 18 that its AI models rewrote parts of AMD chip designs, reporting up to 14% gains in power, performance and area (PPA) on a flagship GPU design. The company said some model-generated changes were included in an AMD design sent for manufacturing, but the figures have not been independently audited and key details about the measured blocks are not public.
OpenAI described four strands of work in a post titled “Advancing chip design with OpenAI models.” The largest figure is an “up to” 14% PPA improvement over what the company called heavily optimised baselines for AMD’s flagship GPU. OpenAI did not identify the blocks behind that best result or report an average across the design. The company also said AMD engineers verified single-digit PPA gains in key blocks of a next-generation flagship CPU.
OpenAI said some GPU changes had entered a design AMD sent for manufacturing, while other changes were added to design files for future versions. The phrase does not establish that the chips have been manufactured, tested or shipped with those changes. OpenAI’s account, as reported by ANI and other outlets, is the basis for those details; the article’s source material says AMD has not published its own figures.
The reported work took place at the register-transfer level (RTL), where engineers describe circuit behaviour in code. OpenAI said its models proposed rewrites to SystemVerilog, and Synopsys design tools checked whether the revised code preserved the original behaviour and measured PPA effects. Those conventional tools, rather than the AI models alone, supplied the reported verification. OpenAI also reported smaller standard-cell area in two units of its own Jalapeño inference chip: about 13% in one unit and 10% in another, with the company saying the changes were validated after the first chips returned.
AI Moves Into Chip Design
The report describes AI being used not just to develop software or assist research, but to propose changes to the hardware that runs AI systems. If the reported gains survive later design checks and manufacturing, they could help chip teams improve efficiency or performance without relying solely on manual redesign. That matters to companies competing to build and operate increasingly demanding AI systems.
However, the figures do not yet show how much a full production chip improves. A best-case result in an unnamed block may have a different practical effect from a similar gain across a large share of a processor. The report also does not establish whether these changes reduce costs, increase shipped performance or affect the availability of AMD products. For now, the news is evidence of a claimed design workflow and reported block-level results, not a verified measure of commercial impact.
Popular right now

Ninja Kitchen System Blender and Food Processor with BlendSense Technology, 1800-Watt, 72 oz, Black (TB401)

Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB-C MagSafe Charging Case

Ninja Foodi Smart Dual Heat Air Fry Oven, 13-in-1, Silver, SP351
As an affiliate, we earn on qualifying purchases.
From RTL Rewrites to Silicon
Chip design proceeds through stages. At RTL, a designer describes what a circuit should do; synthesis tools turn that description into logic, and subsequent placement and routing arrange and connect components for a physical chip. A rewrite can appear beneficial at one stage yet face different constraints after physical implementation. That is why the measurement stage and comparison baseline matter when judging PPA claims.
OpenAI said AMD’s GPU baseline had already been heavily optimised, according to the reports summarising its post. That makes the comparison more demanding than one against an unrefined design, but the public account does not specify the blocks, their share of the chip, or whether the headline gain was measured after synthesis or after placement and routing. OpenAI president Greg Brockman described the work as “roughly an additional year of progress” achieved in less than five months. That is a company characterization of the pace, not an independently measured timeline.
The relationship also has a financial dimension. The source material says OpenAI holds a warrant for up to 160 million AMD shares, issued in October 2025 and tied to deployment of AMD chips. That connection does not disprove the technical results, but it makes independent scrutiny especially relevant when assessing claims made by OpenAI about work benefiting AMD.
“Su praised the results.”
— AMD chief executive Lisa Su, on X, as quoted by Stocktwits
Missing Independent Chip-Level Data
The public information does not identify which AMD GPU blocks produced the 14% best-case figure, how much of the chip those blocks represent, or the average result across tested changes. It is also unclear at which design stage the figure was calculated. Those details are needed to compare a synthesis-stage result with performance in a physically implemented chip.
No independent audit of the reported AMD or Jalapeño results has been published in the supplied material. OpenAI’s account says Synopsys tools checked functional equivalence and measured trade-offs, but tool-based checks do not independently confirm the company’s choice of baseline, test conditions or summary statistics. AMD has not released detailed figures of its own in the cited material. It is also unclear which specific changes are in the design sent for manufacturing and what results that design may produce after fabrication.
OpenAI reported a benchmark involving five models and 9,881 scored runs on two open-source RISC-V processor cores, PicoRV32 and Ibex. GPT-6.2 Astra, released the day before the post, scored highest, according to Inside AI. The source material does not provide enough benchmark detail to infer how that ranking relates to AMD’s commercial chips, so it should be treated as a separate test rather than proof of the GPU gains.
Awaiting AMD and Silicon Results
The next useful evidence would be a technical breakdown from OpenAI or AMD identifying the affected blocks, the baseline and measurement stage, and the distribution of results rather than only the best case. AMD could also clarify which changes entered the design sent for manufacturing and whether it has independently measured their effects.
For the design already sent to manufacturing, later results could show whether the RTL changes translate into the expected physical-chip behaviour. The source material does not provide a timetable for those results or say when AMD may release additional details. Until then, the reported gains remain company-reported engineering results, with their system-wide and commercial effects unresolved.
Key Questions
What did OpenAI report about AMD’s chip designs?
OpenAI said its models rewrote parts of AMD designs and reported up to 14% PPA gains in a flagship GPU design. It also said AMD verified single-digit gains in key blocks of a next-generation flagship CPU.
Does the 14% figure apply to the whole GPU?
The public account does not say that. OpenAI described the result as “up to” 14%, but did not name the blocks measured or give an average across the design.
Who checked whether the AI-generated changes worked?
OpenAI said Synopsys design tools checked that the RTL rewrites preserved the original behaviour and assessed their PPA effects. AMD reportedly verified CPU-block results. No independent audit has been published in the supplied material.
Have the changes reached manufactured AMD chips?
OpenAI said some changes were included in an AMD design sent for manufacturing. That does not confirm that finished chips have been tested or shipped with those changes, and the timing of further results is unclear.
Why does OpenAI’s financial relationship with AMD matter?
The source material says OpenAI holds a warrant for up to 160 million AMD shares, tied to AMD chip deployments. The relationship does not establish that the claims are wrong, but it is relevant when weighing company-reported results and the need for independent evidence.
Source: ThorstenMeyerAI.com
