OpenAI Says Its Models Improved AMD Chip Designs By Up To 14%. What Was Measured, What Wasn’t
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: OpenAI Says Its Models Improved AMD Chip Designs By Up To 14%. What Was Measured, What Wasn’t 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 its models rewrote parts of AMD chip designs, reporting up to 14% improvement across power, performance and area against heavily optimized baselines. The figure is OpenAI’s best reported result, not a disclosed average; AMD has not published its own measurements, and no independent audit is available.

OpenAI said on November 18 that its models had rewritten parts of AMD chip designs, producing up to 14% improvement in power, performance and area against heavily optimized baselines. The company said some model-generated changes were included in an AMD design sent for manufacturing, but it has not disclosed which chip blocks produced the top result or how representative it is.

The reported work focused on register-transfer level (RTL) code, which describes how chip circuits operate. OpenAI’s models proposed alternative RTL implementations, while Synopsys design tools checked that the rewrites preserved the original behavior and evaluated their effects on power, performance and area, according to OpenAI’s account as reported by ANI and other outlets. The model did not independently certify that its own code was correct.

OpenAI described several results that should not be conflated. For an AMD flagship GPU design, it reported a best-case improvement of up to 14% across the three PPA measures, compared with what it called heavily optimized baselines. For key blocks in AMD’s next flagship CPU, OpenAI said AMD had verified single-digit percentage gains. It also said some changes had entered a design sent for manufacturing, while other changes were placed in files for future designs.

The company separately reported results for its own Jalapeño inference chip: model-proposed changes reduced standard-cell area by about 13% in one unit and 10% in another, with the changes validated after the first chips came back, OpenAI said. Reports differ on the name of the first unit, describing it as a warp engine or a vector unit. Those figures concern area in particular units, not a 13% or 10% reduction across the whole chip.

At a glance
reportWhen: Announced November 18, 2026
The developmentOpenAI reported that model-generated changes improved power, performance and area by up to 14% in parts of AMD chip designs, with some changes reportedly included in a design sent for manufacturing.

What Chip Teams Could Gain

If the results hold up, AI-assisted RTL work could help chip engineers test more design alternatives and improve power, performance and area without changing what a circuit is meant to do. Those trade-offs shape how fast a processor runs, how much energy it uses and how much silicon it needs. Improvements can matter for manufacturers and for companies operating large AI systems, where both chip capability and power consumption affect deployment.

The reported manufacturing step gives the announcement practical significance: OpenAI says some generated changes made it into a design sent for fabrication, rather than remaining only in a software benchmark. But the available reporting does not establish the size of those particular changes or whether they will yield measurable benefits in finished products. “Up to 14%” is a best-case figure, and the missing block-level and aggregate data prevent readers from judging the typical gain.

The announcement also illustrates a commercial feedback loop: models are being applied to the hardware used to run AI. That possibility merits attention, but the evidence here remains largely company-reported. OpenAI and AMD have a substantial business relationship, so independent validation would help clarify how much of the reported improvement is attributable to the models and how broadly it applies.

Products Worth Considering

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

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

As an affiliate, we earn on qualifying purchases.

From RTL Code to Silicon

RTL is an early, important stage of chip design. Engineers use it to describe circuit behavior; synthesis software converts that description into logic gates, followed by steps that arrange and connect those gates for manufacturing. A change that looks better in RTL may not retain the same advantage after later design stages, which is why the precise measurement point matters.

OpenAI said its models proposed rewrites that changed logic or how calculations were divided across pipeline stages. For its Jalapeño chip, Inside AI reported examples including a multiplier redesign that used about 15% less cell area in synthesis and reduced latency from two clock cycles to one, and an adder redesign that cut cell area by about 15% at the same two-cycle latency. These are examples for specific units, not the AMD-wide 14% result.

OpenAI also reported testing five models on two open-source processor cores, PicoRV32 and Ibex, across 9,881 scored runs. The source material says GPT-6.2 Astra scored highest in that test. These benchmark results are separate from the claimed improvements to AMD designs and do not independently verify them.

“Su praised the results on X.”

— AMD chief executive Lisa Su, as quoted by Stocktwits

The Missing Measurement Details

The public account does not identify the AMD GPU blocks behind the 14% best case, the share of the chip those blocks represent, or the average result across tested changes. It also does not specify the exact design stage used for the reported GPU measurement, such as whether it was taken after synthesis or after placement and wiring. Without those details, the figure cannot be treated as a whole-chip improvement or compared cleanly with other chip-design results.

OpenAI’s figures have not been independently audited in the material available here, and AMD has not published its own detailed measurements. OpenAI said Synopsys tools checked behavioral equivalence and evaluated trade-offs, but the reported information does not include the underlying test data or enough detail to reproduce the results. The difference between reported gains in design files and final performance in manufactured chips also remains unknown.

There is a financial relationship to keep in view: 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 relationship does not show that the design claims are inaccurate, but it is relevant when evaluating claims made by the companies involved. The source material also says the request for OpenAI’s announcement post was refused; details were reported through ANI, The Decoder and Inside AI.

Evidence From Manufactured Designs

The next useful evidence would be a fuller breakdown from OpenAI or AMD: which blocks were changed, what baseline and design stage were used, how many candidate rewrites were evaluated, and whether the gains persisted through later implementation steps. Results from manufactured chips could show whether the reported design-level improvements translate into measured operating benefits.

AMD’s statements or technical disclosures could also clarify what it independently verified, particularly for the CPU blocks and the design reportedly sent for manufacturing. Until either company publishes more detail, the 14% figure remains an OpenAI-reported best case, while the scale of the gains in commercial products is not established.

Key Questions

What does OpenAI’s “up to 14%” figure measure?

OpenAI said the figure represents a best-case improvement in power, performance and area for parts of an AMD flagship GPU design, compared with heavily optimized baselines. It did not disclose the average result or the specific blocks behind the maximum.

Did the models design a complete AMD chip?

No. The reported work involved rewriting parts of the chip’s RTL code, which describes circuit behavior. The source material does not say that a model designed a complete chip from scratch.

How were the proposed changes checked?

OpenAI said Synopsys design tools checked that the rewrites preserved the original behavior and assessed their power, performance and area trade-offs. The model itself was not the sole verifier.

Has AMD independently confirmed the 14% result?

The source material says OpenAI reported up to 14% for the GPU design and reported AMD-verified single-digit gains in selected CPU blocks. AMD has not published its own detailed figures for the 14% result, and no independent audit is cited.

Does the result mean AMD’s finished chips will be 14% better?

That has not been established. The figure applies to reported design results for unspecified blocks; the source material does not show that a complete manufactured chip will deliver a 14% improvement.

Source: ThorstenMeyerAI.com

You May Also Like

Expanding Our Support For Scientists – Anthropic

Anthropic announces expansion of support for scientists, but details on resources, eligibility, and timing remain undisclosed, raising questions about practical impact.

What Comes After The Apology? DWTS’s Guillermo Rodriguez And Jackson Olson Weigh In

DWTS contestants Guillermo Rodriguez and Jackson Olson say they are friends after Olson apologized for a social media comment about Rodriguez’s dancing.

Team USA Dominates Canada in Basketball Showdown

Dive into Team USA's dominant performance against Canada in basketball, featuring standout players and a preview of their journey to the Paris Olympics.

In Indonesia, Prabowo’s $14bn village co-op drive collides with rural realities

Indonesia’s $14 billion village cooperative initiative led by Prabowo encounters practical obstacles in rural areas, highlighting tensions between policy ambitions and local realities.