OpenAI’s AMD Chip Design Results: A Guide To The Measured Metrics
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: OpenAI’s AMD Chip Design Results: A Guide To The Measured Metrics on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
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 up to 14% gains in power, performance and area against heavily optimized baselines. The headline figure is a best-case result from OpenAI; AMD has not published its own detailed measurements, and independent verification is not available.

OpenAI said November 18 that its models rewrote parts of an AMD flagship GPU design, with the best changes improving power, performance and area (PPA) by up to 14% against heavily optimized baselines. The announcement matters because it points to AI tools being used not only to develop AI software, but also to alter the chips that run it; however, the results are company-reported, and detailed measurements have not been published for independent review.

OpenAI published the results in a post titled “Advancing chip design with OpenAI models.” The available figures in the source reporting come from OpenAI’s account, as summarized by ANI, The Decoder and Inside AI. The company said some model-generated changes were included in a recent AMD design sent for manufacturing, while other changes went into design files for future versions. It did not identify the affected GPU blocks or say what share of the chip they represent.

OpenAI also reported that AMD had verified single-digit percentage PPA gains in key blocks of its next flagship CPU. That statement is attributed to OpenAI; AMD has not published its own detailed figures. Separately, OpenAI said model-proposed changes reduced standard-cell area in two units of its Jalapeño inference chip by about 13% and 10%. The source accounts use different names for one unit, so its exact label is not consistent across reports.

The work focused on register-transfer-level (RTL) code, which describes a chip’s circuit behavior. OpenAI said Synopsys design tools checked whether rewrites preserved the original behavior and evaluated their PPA effects. The reports also describe a test on two public RISC-V processor cores, PicoRV32 and Ibex: five OpenAI models produced 9,881 scored runs, with GPT-6.2 Astra, released the day before the announcement, scoring highest. That benchmark is separate from the AMD chip results and should not be treated as an independent audit of them.

At a glance
reportWhen: Announced November 18, 2026; details re…
The developmentOpenAI reported that its models produced chip-design changes with up to 14% better power, performance and area metrics on an AMD flagship GPU design.

What Chip Teams Can Learn

If the reported gains hold through later design stages and manufacturing, AI-assisted RTL work could help chip teams improve efficiency without changing a chip’s intended behavior. Power, performance and area are linked engineering trade-offs: reducing area can lower manufacturing costs, while power and performance changes affect how a processor runs and what cooling or energy it requires. A useful improvement in one design block may therefore matter to both chipmakers and customers running large computing workloads.

The reported status varies by result. OpenAI said some AMD GPU changes had reached a design sent for manufacturing, but it did not report silicon measurements from that AMD design. The Jalapeño area reductions, by contrast, were said to have been validated after the first chips came back. Those are distinct stages of evidence. The headline “up to 14%” is also a maximum, not an average across a full GPU, and the available material gives no chip-wide figure.

The commercial relationship adds a reason to read the announcement carefully. Since October 2025, OpenAI has held a warrant for up to 160 million AMD shares, which vests as AMD chips are deployed. That arrangement does not establish whether the technical claims are accurate, but it is relevant context for readers evaluating a result promoted by one party with a financial interest in the other’s chip deployment.

Products Worth Considering

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

DJI Mini 4K Drone Combo with 4K Camera, 3-Axis Gimbal, and Two Batteries

DJI Mini 4K Drone Combo with 4K Camera, 3-Axis Gimbal, and Two Batteries

As an affiliate, we earn on qualifying purchases.

From RTL to Manufactured Chips

Chip development moves through several stages. At the RTL stage, engineers describe circuit behavior in code. Synthesis tools convert that description into logic gates; later steps place and connect those gates on a physical chip. A change that looks promising in RTL or synthesis still has to survive subsequent design checks, and manufacturing results can differ from early estimates.

OpenAI’s account describes models proposing alternative implementations of the same function, with conventional design tools checking behavior and estimating trade-offs. Inside AI reported examples from Jalapeño: a multiplier redesign that used about 15% less cell area in synthesis while reducing latency from two clock cycles to one, and an adder redesign that reduced cell area by about 15% at the same two-cycle latency. These examples concern OpenAI’s own chip and are not the same measurement as the AMD GPU’s reported PPA result.

OpenAI president Greg Brockman described the AMD work as roughly a year of progress achieved in less than five months. That is his characterization of the project’s pace, not a published, independently measured comparison of engineering time. OpenAI also said it began using GPT-4o on Jalapeño in early 2025.

The Missing Measurement Details

The public reporting does not specify which AMD GPU blocks produced the reported gains, how large those blocks are relative to the full chip, or whether the 14% figure was measured after synthesis or after physical placement and wiring. It also does not state the average result across all tested changes. Because the figure is described as “up to” 14%, it represents a reported best case, not a typical or whole-chip improvement.

No independent audit of the AMD results has been published in the source material. OpenAI says conventional Synopsys tools checked functional equivalence and evaluated PPA, but the detailed test setup and underlying measurements are not available for outside examination. It is also not clear which changes in the design sent for manufacturing will appear in commercial products, or what measured performance those products will deliver.

The source accounts do not fully align on the name of one Jalapeño unit. They also provide differing levels of detail across the GPU, CPU, custom-chip and open-source-core results. These should be treated as separate claims with different evidence, rather than combined into a single measure of how much AI improves chip design.

Evidence Needed After the Announcement

The next meaningful evidence would be detailed measurements from OpenAI or AMD that identify the affected blocks, baselines, test conditions and design stage, alongside results across more than the best-performing changes. For changes already included in a design sent for manufacturing, later silicon measurements could show whether estimated improvements carried through to fabricated hardware. The companies have not announced a public date for such data.

AMD could also publish its own account of the CPU and GPU results, including which gains it verified and how it measured them. Until that information or an independent assessment becomes available, the announcement establishes what OpenAI says its models achieved—not a public, independently confirmed performance gain across AMD’s chips.

Key Questions

What did OpenAI report about AMD’s GPU design?

OpenAI reported up to 14% improvement in power, performance and area against heavily optimized baselines. It did not publish an average result or identify the design blocks measured.

Did AMD independently publish or confirm the figures?

OpenAI said AMD verified single-digit PPA gains in key blocks of a next-generation flagship CPU. AMD has not published detailed figures of its own in the source material, and no independent audit of the results is reported.

Were the changes tested on physical chips?

OpenAI said some AMD GPU changes were included in a design sent for manufacturing, but it did not provide measurements from that AMD silicon. The reported Jalapeño area reductions were said to have been validated after OpenAI’s first chips came back.

What does PPA mean in this announcement?

PPA means power, performance and area: three measures chip designers balance when changing a circuit. A reported gain in a particular block does not by itself establish the same improvement across an entire processor.

Source: ThorstenMeyerAI.com

You May Also Like

Will Geo-Trade Shifts Play A Role In Francesca Hong’s Wisconsin Gubernatorial Bid?

Analysis of how geopolitical trade developments may influence Francesca Hong’s 2026 Wisconsin gubernatorial campaign and primary prospects.

Exploring The Trend: AI Operations Resembling Data Center REITs

Emerging trend shows AI operations now resemble data center REITs, raising questions about infrastructure and strategic focus for AI deployment.

Trade voice copilo

Trade voice copilo is being tested as a workflow tool for small trades businesses to streamline job notes and invoicing using speech-to-text and AI integration.

Trade Conflict Deepens: Canada’s Supply Chain Prepared For US Tariffs

Canada plans to match US tariffs dollar-for-dollar if trade negotiations break down, signaling a deepening trade conflict and potential supply chain disruptions.