AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: OpenAI’s Models In AMD Chip Design: A Measured Gain, Not The Whole Story on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

OpenAI said on November 18 that its models rewrote parts of AMD chip designs, reporting best-case improvements of up to 14% in power, performance and area for a flagship GPU. The company says some changes reached a design sent for manufacturing, but the figures, measurement details and independent verification remain limited.

OpenAI said November 18 that its models rewrote sections of AMD chip designs, reporting up to 14% gains in power, performance and area on a flagship GPU design. OpenAI says AMD has included some model-generated changes in a design sent for manufacturing, but the headline figure is a best-case result from the companies’ account, with no independent audit published.

The work focused on register-transfer level (RTL) code, which describes a chip’s circuit behavior. OpenAI’s models proposed alternative SystemVerilog designs; according to the company, Synopsys design tools checked that the rewrites preserved the original behavior and evaluated their power, performance and area trade-offs. Conventional verification tools, rather than the model alone, performed the checks.

OpenAI reported “up to” 14% improvement in PPA for AMD’s flagship GPU against what it described as heavily optimized baselines. The company also said AMD had verified single-digit percentage gains in key blocks of its next flagship CPU. OpenAI did not publish the names or sizes of the affected blocks, nor an average gain across the designs.

The announcement included results from OpenAI’s Jalapeño inference chip. The company said model-proposed changes reduced standard-cell area by about 13% in one unit and 10% in another, with validation after the first chips returned. Reports differ on the name of one unit: ANI calls it a vector unit, while Inside AI describes it as a warp engine. OpenAI also reported a test of five models on two open-source RISC-V processor cores, PicoRV32 and Ibex, totaling 9,881 scored runs; according to Inside AI, GPT-6.2 Astra achieved the highest score.

At a glance
reportWhen: Announced November 18, 2026
The developmentOpenAI reported that its models helped optimize parts of AMD chip designs, including a flagship GPU and a forthcoming CPU.

AI Models Enter Chip Design Work

The report offers a concrete example of AI being applied not only to software and research, but to the hardware used to run AI systems. If changes survive later design and manufacturing stages, reductions in area or gains in performance could affect chip capacity, power use and production costs. The announcement does not establish that the reported improvements will translate into equivalent benefits across a complete processor or in deployed systems.

It also shows why verification and the quality of the comparison matter. A circuit rewrite must retain the design’s behavior; a subtle error can undermine a chip and may require costly correction. OpenAI says Synopsys tools checked equivalence and evaluated design trade-offs, but the public account does not provide enough detail for outside engineers to reproduce or independently assess the AMD results.

There is also a commercial relationship readers should account for. Since October 2025, OpenAI has held a warrant for up to 160 million AMD shares, which vests as AMD chips are deployed. That connection does not disprove the engineering results, but it makes clear why independently available measurements would help readers judge the claims.

Products Worth Considering

Jackery Explorer 2000 v2 Portable Power Station, 2042Wh, 2200W

Jackery Explorer 2000 v2 Portable Power Station, 2042Wh, 2200W

As an affiliate, we earn on qualifying purchases.

From RTL Rewrites to Fabrication

RTL is an early stage of chip design: engineers describe how circuit blocks behave, and synthesis tools translate that description into logic gates. Further steps place and connect those gates on silicon. OpenAI’s account concerns rewriting RTL, not redesigning every stage of chip production. A gain at one point in the process may change as a design moves through later implementation steps.

OpenAI compared the AMD GPU work with heavily optimized baselines, which the company characterized as designs AMD engineers had worked on for years. That is a more demanding comparison than starting from an unoptimized design. Still, “up to 14%” identifies the best reported result, not the typical result, and the announcement as relayed by ANI, The Decoder and Inside AI does not specify the affected blocks or provide a full breakdown.

OpenAI president Greg Brockman described the work as “roughly an additional year of progress on @AMD’s chip achieved in less than 5 months using our models,” according to Stocktwits. OpenAI’s post, titled “Advancing chip design with OpenAI models,” appeared November 18. The source account says an automated request to the post was refused, so the reported details rely on coverage by other outlets rather than direct review of the full post.

The Missing Measurement Details

The public account does not identify the GPU blocks involved, how much of the chip they represent, or whether the reported gains were measured after synthesis or after placement and wiring. It also does not state the typical improvement, give full measurement data, or explain how the results varied between design attempts. The 14% figure should not be read as a whole-chip improvement.

OpenAI says some changes entered a design AMD sent for manufacturing, while others went into production design files for later versions. It is not clear from the available material which specific changes are in the manufacturing submission, what the final silicon results will be, or whether the claimed gains will persist in products. No independent audit of the chip-design results has been published, and AMD has not released its own figures.

The Jalapeño results provide another example, but the differing names used for one unit in reports leave a terminology discrepancy. The RISC-V benchmark is more open to scrutiny because the processor cores are public, yet the supplied information does not include enough methodology or score detail to compare those tests fully with AMD’s proprietary designs.

Silicon Results and Independent Checks

The next meaningful evidence will come from later design stages and, where available, measurements from manufactured chips. Those results could show whether the RTL-level changes retained their claimed power, performance and area benefits after implementation. OpenAI has not provided a timeline for publication of that data in the material available here.

More detailed disclosures from AMD or OpenAI could clarify which blocks changed, the baseline and measurement stage for each result, and whether gains were typical or limited to selected cases. Independent verification or reproducible benchmark data would also help distinguish company-reported results from independently confirmed performance. Until then, the announcement establishes that the companies say model-assisted redesign has progressed into chip-design workflows, not that a 14% improvement applies across AMD’s flagship GPU.

Key Questions

What did OpenAI say its models changed?

OpenAI said its models rewrote parts of chip designs in SystemVerilog RTL, the code describing circuit behavior. It reported results for AMD GPU and CPU designs and for its own Jalapeño inference chip.

Does the 14% figure apply to the entire AMD GPU?

No. OpenAI described it as an “up to” result against heavily optimized baselines, but did not name the affected blocks or say what share of the GPU they represent. The available information does not support treating it as a whole-chip gain.

Who checked whether the rewritten circuits worked?

OpenAI said Synopsys design tools verified that the changes preserved the original behavior and assessed power, performance and area. The public account does not include an independent audit of the AMD results.

Have the changes reached manufacturing?

OpenAI said some model-generated changes were included in a design AMD sent for manufacturing. It has not specified which changes, and the material available does not report final silicon measurements confirming their effect.

Why is OpenAI’s relationship with AMD relevant?

Since October 2025, OpenAI has held a warrant for up to 160 million AMD shares, which vests as AMD chips are deployed. That financial connection is relevant when evaluating company-reported results; it does not by itself determine whether the engineering claims are accurate.

Source: ThorstenMeyerAI.com

You May Also Like

Comparing 10 Mobile Workstation Laptops For Professional AI Work In 2027

A source-based comparison of 10 mobile workstations highlights GPU, memory, portability and thermal tradeoffs for professional AI and engineering work.

How Chatham Is Applying OpenAI To Capital Markets Expertise

OpenAI published an announcement about Chatham and capital markets expertise, but the available details do not identify tools, rollout scope or results.

The Economic Impact Of Claude Opus 5.5 On AI Deployment

Anthropic’s Claude Opus 5.5 boosts AI efficiency and cuts costs, influencing AI deployment strategies amid industry competition.

Claude AI Finds A Linux Flaw That Could Let Hackers Reach Root Access

A How-To Geek headline says Claude found a Linux flaw that could allow root access, but the vulnerability and affected systems are not identified.