What The 14% AMD Chip Design Improvement Does—and Doesn’t—Tell Us
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: What The 14% AMD Chip Design Improvement Does—and Doesn’t—Tell Us on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

  • 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 its models rewrote parts of AMD chip designs, reporting up to 14% improvement in power, performance and area against heavily optimized baselines. The figure is a best-case company-reported result; the affected blocks, average gains and independent verification have not been disclosed. AMD reportedly verified single-digit gains in key CPU blocks, while conventional design tools checked that proposed rewrites preserved behavior.

OpenAI said on November 18 that its models rewrote parts of designs for AMD’s flagship graphics processor, producing up to 14% gains in power, performance and area, or PPA, compared with heavily optimized baselines. OpenAI also said some model-generated changes had been included in an AMD design sent for manufacturing, but it has not disclosed which blocks produced the headline result or published an independent audit.

The announcement describes several results, not one uniform improvement across an entire chip. OpenAI reported up to 14% better PPA on AMD’s flagship GPU design. Separately, it said AMD had verified single-digit percentage gains in key blocks of its next flagship CPU. OpenAI said some changes were in a recent design sent for manufacturing, while other proposed changes were added to AMD’s design files for future versions. The company did not identify the affected blocks or give an average result.

OpenAI also reported results from its own Jalapeño inference chip. Model-proposed changes reduced standard-cell area by about 13% in one unit and 10% in another, according to the company. Reports differ on the name of the first unit: ANI described it as a vector unit, while Inside AI called it a warp engine. OpenAI said it validated these results when the first chips came back.

The models worked on register-transfer level (RTL) code, which describes chip behavior cycle by cycle. OpenAI said Synopsys design tools checked that each rewrite preserved the original behavior and evaluated its PPA trade-offs. This verification was performed by conventional engineering software, rather than by the models themselves. The distinction matters because a change that improves speed or area but alters a chip’s behavior cannot be treated as a successful design improvement.

At a glance
reportWhen: Announced November 18, 2026; details re…
The developmentOpenAI reported that its models improved selected AMD chip-design blocks by up to 14% on power, performance and area, with some changes already in a design sent for manufacturing.

AI Rewrites Could Shorten Chip Design

If the reported improvements hold through later design stages and manufacturing, the work could help engineers improve chips without changing what those chips are meant to do. Better power, performance and area can affect operating costs, computing capacity and the space available for other components. The possibility of models assisting with chip design also connects AI development to the hardware that future AI systems will use.

The manufacturing detail is a meaningful step beyond a software-only demonstration: OpenAI said some changes were included in a design AMD sent for manufacturing. But that does not establish that the manufactured part has delivered the claimed gains in a finished product. Chip designs pass through multiple stages after RTL, and results can change as logic is synthesized, placed and wired. The announcement offers a potential design-process benefit, not yet a public measurement of production-chip performance.

For readers and customers, the relevant question is not simply whether a model can suggest a smaller or faster circuit. It is whether the change remains correct, survives the rest of the engineering process and yields a repeatable gain. The reported use of Synopsys verification tools addresses part of that question, while the missing block-level data and independent review limit how much can be concluded from the headline percentage.

Breville Barista Express Manual Espresso Machine with Built-In Grinder, Brushed Stainless Steel

Breville Barista Express Manual Espresso Machine with Built-In Grinder, Brushed Stainless Steel

KitchenAid 5.5 Qt Bowl-Lift Stand Mixer – Porcelain White (KSM55SXXX)

KitchenAid 5.5 Qt Bowl-Lift Stand Mixer – Porcelain White (KSM55SXXX)

As an affiliate, we earn on qualifying purchases.

From RTL Rewrites to Silicon

RTL is an early, code-based description of a chip’s logic. Engineers and synthesis tools use it to produce a circuit design, followed by steps that arrange and connect components for fabrication. OpenAI’s account places its models at the RTL stage: they proposed alternative implementations of existing functions, rather than designing and manufacturing a complete processor independently.

OpenAI said the AMD comparison was against heavily optimized baselines, which the company’s account described as designs AMD engineers had optimized over years. That makes the comparison more informative than one against an intentionally simple design. It still leaves important details unavailable, including the specific blocks tested and whether the reported gains were measured after synthesis or after physical placement and wiring.

OpenAI also described tests on two public RISC-V processor cores, PicoRV32 and Ibex. It said five models completed 9,881 scored runs, with GPT-6.2 Astra receiving the highest score. This benchmark is separate from the AMD claims and should not be treated as independent confirmation of them. OpenAI’s announcement was reported by ANI, The Decoder and Inside AI; this article’s figures are based on those reports and the company’s statements.

There is also a financial relationship between the companies. The source material says OpenAI has held since October 2025 a warrant for up to 160 million AMD shares, with vesting tied to deployment of AMD chips. That relationship does not establish that the technical claims are wrong, but it is relevant when weighing company-reported results. The source account also says neither AMD nor OpenAI published an independent audit of the design gains.

The Missing Block-Level Evidence

The public information does not identify which AMD GPU blocks generated the up to 14% figure, how much of the chip those blocks represent, or what the typical result was. It also does not specify whether the reported PPA measurements came after synthesis, physical placement and wiring, or another design stage. Without those details, the percentage cannot be interpreted as a chip-wide improvement.

There is no published independent audit of the AMD results in the supplied reporting, and AMD has not released its own figures. OpenAI’s account says AMD verified single-digit CPU gains, but the scope and test conditions are not public. The reports also provide no product-level results from the design sent for manufacturing, so it remains unclear whether the gains will persist in shipped hardware.

The underlying announcement could not be directly accessed by the source writer, who relied on reports from ANI, The Decoder and Inside AI. Those outlets agree on the key figures, but readers do not have public block-level data or a complete methodology with which to reproduce the claims. The financial relationship between OpenAI and AMD is another reason to distinguish the companies’ statements from independently checked findings.

Awaiting Manufacturing and Methods Data

The next useful evidence would be detailed results from OpenAI or AMD identifying the affected blocks, measurement stage, comparison conditions and distribution of outcomes—not only the best case. Publication of independent technical review or reproducible measurements would help establish whether the gains are consistent and how much of the chip they affect.

The design reportedly sent for manufacturing may provide a further test, but the announcement does not state when results from that process will be available or whether either company plans to publish them. Until such information appears, the 14% figure remains an OpenAI-reported maximum, while the practical effect on finished AMD products is unconfirmed.

Key Questions

What does OpenAI’s 14% figure measure?

OpenAI reported up to 14% improvement in power, performance and area on parts of AMD’s flagship GPU design, against heavily optimized baselines. It did not disclose the specific blocks, the average gain or the measurement stage.

Does the result mean AMD’s whole GPU is 14% better?

No. “Up to” describes a best-case result, and the affected blocks and their share of the chip have not been identified. The available information does not support saying the entire GPU improved by 14%.

Did AMD confirm the reported gains?

OpenAI said AMD had verified single-digit gains in key blocks of its next flagship CPU. The source material says AMD’s chief executive Lisa Su praised the results on X, but AMD has not published its own figures or an independent report.

Were the model-generated chip changes checked for correctness?

According to OpenAI, Synopsys design tools checked whether the rewritten RTL preserved the original behavior and evaluated power, performance and area. That is a reported verification step, not an independent audit of the overall improvement claims.

Have the improvements been demonstrated in products?

OpenAI said some changes were included in an AMD design sent for manufacturing. The public information does not report final product measurements, shipment timing or whether the claimed gains carried through to manufactured hardware.

Source: ThorstenMeyerAI.com

You May Also Like

A Closer Look At Nvidia Flora’s 125× Robot Video Labeling Cost Reduction

Nvidia says Flora could label robot video for $0.72 an hour, but its 125× cost comparison pairs a modeled estimate with measured API spending.

AI’s Unblinking Radar: How It Supports Critical Infrastructure

Exploring how synthetic aperture radar (SAR) satellites, supported by AI, provide persistent, reliable surveillance for critical infrastructure worldwide.

Microsoft Killed FoxPro In 2007. Anyway, Here’s FoxPro Revived

FoxPro, officially discontinued by Microsoft in 2007, appears to be making a comeback amid renewed interest and ongoing legacy use.

Qualcomm Surges In Global Coverage

Qualcomm’s media mentions have surged dramatically, reaching 38 mentions in recent coverage, indicating heightened global attention on the company.