🔍 Read the full analysis: OpenAI’s Models And AMD Chips: Measuring Design Gains Without Overstating Them on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI said on November 18 that its models rewrote parts of AMD chip designs, reporting up to 14% gains in power, performance and area for a flagship GPU and single-digit gains in key blocks of a next-generation CPU. The figures come mainly from OpenAI, which has a financial relationship with AMD; the companies have not released block-level data or an independent audit.
OpenAI said November 18 that its models rewrote parts of circuit designs for AMD processors, reporting up to 14% improvements in power, performance and area (PPA) for an AMD flagship GPU. OpenAI also said some model-generated changes were included in a design AMD sent for manufacturing, but the reported gains have not been independently audited and detailed test results are not public.
The work focused on register-transfer level (RTL) code, which describes how a chip’s circuits behave. OpenAI said its models proposed alternative implementations and that Synopsys design tools checked whether the rewrites preserved the original behavior and measured PPA. The claims were reported by ANI and other outlets summarizing OpenAI’s announcement; the underlying post was not directly accessible to the source article’s author.
OpenAI described several separate results. For AMD’s flagship GPU, it reported up to 14% better PPA against what it called heavily optimized baselines. For key blocks of AMD’s next flagship CPU, OpenAI said AMD had verified single-digit percentage gains. The company also said it incorporated some proposed changes into an AMD design sent for manufacturing, while other changes entered files for future production designs.
OpenAI reported results on its own custom inference chip, Jalapeño, too: model-proposed changes reduced standard-cell area by about 13% in one unit and about 10% in another, according to the company. OpenAI said those results were validated after the first chips returned from manufacturing. The source reports use different names for one of the units, calling it a warp engine or a vector unit, so its exact identification is not consistent across accounts.
AI Rewrites Could Affect Chip Costs
If the reported changes survive later design and manufacturing checks, using AI to improve chip layouts could help manufacturers get more performance from a design, reduce power use, or fit functions into less silicon. Those gains matter because advanced processors require lengthy engineering work and expensive fabrication; even improvements to selected blocks could affect the capabilities or production economics of future chips.
The announcement also points to a potential feedback loop: models may help design the hardware used to train and run later models. But the headline percentage does not establish a whole-chip improvement. OpenAI described a best-case “up to” figure, without identifying the blocks or saying how much of the GPU they represent. A local gain may have a different effect on the finished processor, and the three PPA measures can involve trade-offs.
The relationship between the companies is relevant to how readers should assess the claims. Since October 2025, OpenAI has held a warrant for up to 160 million AMD shares, which vests as AMD chips are deployed, according to the source material. That financial tie does not show the reported results are wrong, but it makes independent evidence particularly useful.
Popular right now

Roborock Qrevo Edge 3 Pro Robot Vacuum and Mop, 36,000 Pa

Ninja Mega Kitchen System Blender and Food Processor, 1500W, 72 oz Pitcher, 8-Cup Bowl, Black (BL770)

Dyson V9 Motorbar Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
From RTL Code to Manufactured Chips
RTL is an early, important stage in chip design. Engineers use it to specify what hardware blocks do; synthesis tools translate that description into logic gates, followed by steps that place and connect components on a chip. OpenAI’s claim concerns changes at the RTL stage, not a general redesign of an entire processor by a model.
In its account, conventional verification tools—not the AI model—checked that proposed changes preserved the circuit’s behavior. That distinction matters: a faster or smaller implementation is not useful if it produces incorrect results. OpenAI described the comparison for AMD as being against designs AMD engineers had already optimized, rather than against an unrefined starting point. However, the public summaries do not provide enough technical detail to reproduce that comparison.
OpenAI also described tests on two open-source processor cores, PicoRV32 and Ibex. It said five models completed 9,881 scored runs, with GPT-6.2 Astra scoring highest. These benchmarks are a separate line of evidence from the AMD results: they concern public cores and model performance on design tasks, not a direct measurement of production AMD chips.
“Su praised the results on X.”
— AMD chief executive Lisa Su, as quoted by Stocktwits
Key Data Behind the Gains Is Missing
The public reporting does not identify which AMD GPU blocks produced the up to 14% result, what fraction of the processor those blocks occupy, or whether the figure reflects synthesis results or a later stage after placement and wiring. It also gives no average across tested changes, making it impossible to tell how typical the best result was.
OpenAI says AMD verified single-digit gains in key CPU blocks, but AMD has not published its own figures in the material available here. No independent audit of the AMD or Jalapeño results is cited. The specific manufacturing status and eventual silicon performance of the changed AMD design are also not established by the announcement. These gaps leave readers unable to compare the claimed improvements consistently or assess their effect on complete chips.
Watch for AMD’s Design Results
The next useful evidence would be a technical account from AMD or OpenAI that identifies the tested blocks, the measurement stage, the baseline and the spread of results—not only the best case. For the design sent for manufacturing, later information could show whether the changes were retained and how they performed in finished silicon.
Until those details are released, the announcement supports a narrower conclusion: OpenAI says its models generated RTL changes that passed conventional behavioral checks and produced measured improvements in selected designs. The size and practical reach of those gains remain uncertain, particularly at whole-chip level.
Key Questions
What did OpenAI say its models changed?
OpenAI said its models rewrote parts of chip designs at the RTL stage, proposing alternative circuit implementations. The company said Synopsys tools checked behavior and measured power, performance and area.
Does the 14% figure apply to the whole AMD GPU?
The public account describes up to 14% improvement against heavily optimized baselines, but does not specify the blocks tested or how much of the GPU they represent. It should not be read as a confirmed whole-chip gain.
Has AMD independently confirmed the reported figures?
OpenAI said AMD verified single-digit gains in key blocks of a next-generation CPU. The source material says AMD has not published its own detailed figures, and no independent audit is cited.
Were any model-generated changes used in a manufacturing design?
OpenAI said some changes were included in an AMD design sent for manufacturing. The announcement, as reported by the cited outlets, does not establish the final silicon results or whether every proposed change remained in the design.
Why does OpenAI’s relationship with AMD matter?
OpenAI holds a warrant for up to 160 million AMD shares that vests as AMD chips are deployed, according to the source material. That tie is relevant disclosure when weighing company-reported results, but it does not by itself establish whether the claims are accurate.
Source: ThorstenMeyerAI.com
