🔍 Read the full analysis: Can AI Improve AMD Chip Designs? OpenAI Reports Gains Of Up To 14% 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 says 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 says some changes are in a design sent for manufacturing, but it has not disclosed the blocks, average results or an independent audit.
OpenAI said November 18 that its models rewrote parts of designs for AMD processors, reporting up to 14% improvement in power, performance and area on a flagship GPU compared with what it called heavily optimized baselines. OpenAI says some model-generated changes are already in an AMD design sent for manufacturing, though the headline figure and implementation details have not been independently verified.
The work targets register-transfer-level (RTL) code, which describes how a chip’s circuit blocks operate. OpenAI says its models proposed alternative RTL implementations, while Synopsys design tools checked that the rewritten circuits preserved the original behavior and assessed their power, performance and area, or PPA. The verification was done with conventional engineering software, not by relying on the model to certify its own work.
OpenAI reported several results that should not be conflated. For the AMD flagship GPU, it described a best-case PPA improvement of up to 14%, without identifying the blocks measured or giving an average. For a next-generation AMD flagship CPU, OpenAI said AMD had verified single-digit PPA gains in key blocks. The company also says AMD included some generated changes in a design sent for manufacturing and placed other changes in design files for future products.
OpenAI separately reported changes to Jalapeño, its own inference chip: standard-cell area fell about 13% in one unit and 10% in another, with the results validated after the first chips returned, according to the company. Reports differ on the name of the first unit, describing it as either a warp engine or a vector unit. These Jalapeño figures are not the same measurement as the AMD GPU’s reported PPA improvement.
AI Reaches Chip Engineering
If the reported results hold up, models could help engineers test more circuit implementations and find useful trade-offs sooner. Chip design involves balancing power use, speed and silicon area; improving one can make another worse. A model that proposes functionally equivalent alternatives could shorten parts of that search, while engineering tools and human teams remain responsible for checking whether the designs work.
The manufacturing detail makes the announcement more consequential than a software-only demonstration: OpenAI says some changes have moved into a design sent for fabrication. But that does not establish that the chips have been manufactured, shipped or shown to deliver the reported gains in a finished product. The reported up-to figure is a best case, and without block-level data or a typical result, readers cannot tell how much of an overall chip benefits.
The commercial relationship also matters when weighing the claims. OpenAI says it has a warrant for up to 160 million AMD shares, vesting as AMD chips are deployed. That arrangement does not disprove the technical results, but it is relevant context: OpenAI has a financial interest tied to AMD deployments, and the public figures so far come largely from OpenAI rather than an independent assessment.
Products Worth Considering

Apple MacBook Air 13-inch with M5 Chip, 16GB Memory, 512GB SSD, Liquid Retina Display, and 12MP Center Stage Camera – Midnight

Dell UltraSharp U2424H 24-Inch FHD Monitor

Apple Mac mini Desktop Computer with M5 Pro Chip (2026)
As an affiliate, we earn on qualifying purchases.
From RTL to Fabrication
RTL is an early, code-based representation of a chip’s behavior. Engineers describe what circuit blocks do, and synthesis software converts that description into logic gates. Later design stages arrange and connect those gates on silicon. OpenAI’s announcement concerns rewriting RTL, not a claim that its models independently designed a complete processor or replaced the rest of the chip-development process.
According to OpenAI, the models explored alternatives such as changing logic, sharing hardware, adjusting how calculations are divided across pipeline stages, or reducing intermediate storage. Reports on the announcement describe Jalapeño 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% while keeping the same two-cycle latency. Those examples concern specific blocks and do not establish whole-chip gains.
OpenAI also described testing five of its models on two open-source processor cores, PicoRV32 and Ibex, across 9,881 scored runs. Its account says GPT-6.2 Astra, released the day before the announcement, scored highest. That benchmark is separate from the AMD design results. The announcement was published as “Advancing chip design with OpenAI models” on November 18, 2026; the reported figures here are based on coverage by ANI, The Decoder and Inside AI because the supplied source says an automated request to OpenAI’s post was refused.
“The source material says Su praised the results but does not provide her exact wording.”
— AMD chief executive Lisa Su, in a post on X, as quoted by Stocktwits
Key Figures Still Unverified
Important measurement details remain undisclosed. OpenAI has not identified the GPU blocks behind the 14% result, said what portion of the chip they represent, or reported an average across the work. The source accounts also do not specify whether the figures were measured after synthesis or after placement and wiring, later stages that can change the final outcome.
The public material does not provide a full account of the test conditions, the distribution of results, or independent confirmation of the reported AMD gains. It is also unclear which changes are in the design sent for manufacturing, what stage that design has reached, and whether the reported gains will persist in manufactured products. No independent audit of the chip-design results is described in the source material.
The financial relationship adds a reason to seek outside scrutiny, but it is not evidence that the reported measurements are false. AMD’s own detailed figures and a block-by-block account have not been published in the material provided. The difference between a best-case result and typical performance therefore remains unresolved.
Awaiting AMD and Silicon Results
The next useful evidence would include AMD’s own technical figures, details on which blocks changed, and a clear account of the measurement stage and baseline. Reporting on whether the design sent for manufacturing completes fabrication, and whether the changes produce the expected results in silicon, could test the announcement beyond RTL and design-tool measurements.
OpenAI says some changes are in future AMD production design files, but it has not provided a schedule for those products or a date for public performance results. Until AMD, OpenAI or an independent evaluator releases more detail, the reported gains remain company-reported results rather than a verified measure of improvement across an entire chip.
Key Questions
What did OpenAI report about AMD’s GPU?
OpenAI reported up to 14% improvement in power, performance and area on parts of an AMD flagship GPU design, compared with heavily optimized baselines. It did not disclose the measured blocks or an average result.
Does the 14% figure apply to the whole GPU?
The available account does not say that it does. OpenAI described a best-case result but did not identify the blocks or state what share of the chip they represent. The figure should not be read as a confirmed whole-chip improvement.
How were the proposed chip changes checked?
OpenAI says Synopsys design tools checked that rewrites preserved the original circuit behavior and assessed their PPA effects. The model proposed alternatives; conventional verification tools performed the described checks.
Have the changes been manufactured?
OpenAI says some model-generated changes are included in an AMD design sent for manufacturing. The source material does not confirm that the resulting chips have been manufactured, shipped or independently tested.
Has AMD independently confirmed the reported figures?
OpenAI says AMD verified single-digit PPA gains in key blocks of a next-generation flagship CPU, and AMD chief executive Lisa Su praised the results on X. The source material says AMD has not published its own detailed figures, and it describes no independent audit of the results.
Source: ThorstenMeyerAI.com
