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

🔍 Read the full analysis: OpenAI And AMD: Which Chip Design Outcomes Did The AI Models Improve? 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 and produced up to 14% improvements in power, performance and area against heavily optimized baselines. The company also reported single-digit gains in selected CPU blocks and area reductions in its own chip, but independent audits and several measurement details have not been published.

OpenAI said November 18 that its AI models rewrote parts of designs for AMD processors, reporting up to 14% gains in power, performance and area (PPA) on an AMD flagship GPU and single-digit gains in selected blocks of a forthcoming CPU. OpenAI says AMD has included some model-generated changes in a design sent for manufacturing, but the results have not been independently audited and the companies have not disclosed which GPU blocks produced the highest figure.

The results were described in an OpenAI post titled “Advancing chip design with OpenAI models.” OpenAI’s figures, reported by ANI, The Decoder and Inside AI, refer to models rewriting SystemVerilog register-transfer-level (RTL) code—a description of how chip circuits behave. The company said the GPU comparison was against heavily optimized baselines, rather than an unoptimized design. “Up to 14%” identifies a best reported result; OpenAI did not publish an average or say how much of the GPU the relevant blocks represent.

OpenAI also said AMD had verified single-digit percentage PPA gains in key blocks of its next flagship CPU. For its own custom inference chip, called Jalapeño, OpenAI reported standard-cell area reductions of about 13% in one unit and 10% in another. The descriptions differ across reports: ANI calls the first unit a vector unit, while Inside AI calls it a warp engine. OpenAI said the Jalapeño changes were validated after the first chips came back.

For checking, OpenAI said Synopsys design tools verified that proposed RTL changes preserved the original behavior and measured their PPA effects. The model proposed rewrites; conventional design tools checked them. OpenAI also reported a separate test of five models on two open-source RISC-V processor cores, PicoRV32 and Ibex, with 9,881 scored runs. Its newly released GPT-6.2 Astra ranked highest in that test, according to Inside AI. Those benchmark results are distinct from the AMD design claims.

At a glance
reportWhen: Announced November 18, 2026; further te…
The developmentOpenAI reported that its AI models improved selected AMD chip-design outcomes by rewriting circuit descriptions, with some changes reportedly included in a design sent for manufacturing.

AI Rewrites Could Shorten Chip Design

If the reported gains hold across production designs, AI-assisted RTL work could help chipmakers improve performance, power use or silicon area without starting from scratch. These measures compete: a change that improves speed may consume more power or area, so a useful result depends on the trade-offs and on whether the design remains correct.

The reported manufacturing step gives the announcement more practical weight than a demonstration limited to code or simulation. OpenAI says some changes were included in an AMD design sent for manufacturing, while other changes entered files for future production designs. However, the announcement does not establish that the reported maximum will appear in a shipping product, or quantify any effect on product performance, energy consumption, cost or release timing.

The relationship between the companies also matters when readers weigh the claims. OpenAI holds a warrant for up to 160 million AMD shares, issued in October 2025, that vests as AMD chips are deployed. That financial connection does not show that the engineering results are wrong, but it makes independent evidence and clear measurement details particularly relevant.

Products Worth Considering

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

As an affiliate, we earn on qualifying purchases.

From RTL Code to Silicon

RTL is an early stage of digital chip design. Engineers use code to specify a circuit’s behavior cycle by cycle; synthesis software converts that description into logic gates, and later processes arrange and connect those gates for fabrication. OpenAI’s reported work targets the RTL stage, where engineers may try alternative logic, pipeline arrangements or ways to share hardware.

OpenAI described a benchmark alongside the company-specific results. It tested five models on the open-source PicoRV32 and Ibex processor cores, totaling 9,881 scored runs; GPT-6.2 Astra scored highest, according to Inside AI. This provides a separate test setting using public designs, but it does not independently verify the claimed improvements to AMD products.

OpenAI president Greg Brockman characterized the AMD work as “roughly an additional year of progress” achieved in less than five months. That is a company representative’s description of the pace of work, not a published independent measurement of development time saved. The underlying announcement was published November 18, 2026; the available reports attribute the detailed technical figures to OpenAI.

“Su praised the results on X.”

— Lisa Su, AMD chief executive, as quoted by Stocktwits

Key Metrics Still Undisclosed

The public reporting does not name the AMD GPU blocks associated with the 14% maximum, say what portion of the chip they occupy, or provide the average result across changes. It also does not specify whether the reported gains were measured after synthesis or after placement and wiring, later stages that can alter chip-level outcomes. The baseline is described as heavily optimized, but the detailed designs and measurement procedures are not public in the source material.

No independent audit of the AMD or Jalapeño results has been published. OpenAI says Synopsys tools checked functional equivalence and measured trade-offs, but the reported materials do not provide an external review of the complete process or reproducible results. AMD has not published its own figures, and the exact CPU blocks and size of their reported gains are not specified. It is also unclear which announced changes will reach a finished product, or when.

Watch for AMD Design Details

The next useful evidence would be AMD-published measurements identifying the affected blocks, baseline, design stage and trade-offs, alongside detail on which changes made it into manufacturing. Independent review or reproducible technical documentation could help establish how broadly the results apply. OpenAI’s announcement says some changes are already in a design sent for manufacturing and others are in files for future designs, but neither company has provided a public schedule for additional disclosures or product-level results in the reported material.

Key Questions

What did OpenAI say its models improved?

OpenAI reported up to 14% PPA improvement in parts of an AMD flagship GPU, single-digit gains in selected blocks of an AMD flagship CPU, and smaller standard-cell area in two units of its Jalapeño inference chip.

Does the 14% figure describe the whole AMD GPU?

No. OpenAI reported an “up to” result against heavily optimized baselines, but did not name the blocks, say what fraction of the GPU they represent, or provide an average. The figure should not be read as a whole-chip improvement.

How were the proposed circuit changes checked?

OpenAI said Synopsys design tools checked that the RTL rewrites preserved the original behavior and assessed their power, performance and area effects. No independent audit of the reported results has been published.

Have the changes reached manufacturing?

OpenAI said AMD included some model-generated changes in a design sent for manufacturing, with other changes placed in files for future production designs. The reports do not identify the specific changes or establish what will appear in a finished product.

Source: ThorstenMeyerAI.com

You May Also Like

The Storage Challenges Behind ChatGPT’s Rapid User Growth (Part II)

OpenAI’s article title frames an online-storage effort for more than 1 billion ChatGPT users, but technical details and the figure’s basis are unavailable.

Transcription, Translation, And Attribution Get Faster With Granite Speech 5.0

IBM announces a multilingual 2B model and updates to Turbo CTC and Granite Speech 4.1, citing gains in speed, transcription and speaker labeling.

Shop For Small Business AI Automation Software With This Prime Big Deal Days Guide

A small-business guide outlines AI automation costs, starter tasks and risks, but provides no Prime Big Deal Days offers or confirmed event dates.

A Look At 12 AI Voice Cloning Software Guides For 2027

A ThorstenMeyerAI.com comparison covers 12 voice-cloning guides, but the titles are learning resources, not software products.