Separating AI hype from reality.

OpenAI Says Its Models Cut Chip Design Time, With Gains on AMD GPU and Own Jalapeño Chip

November 19, 2026 · OpenAI · AMD · Synopsys

OpenAI published a research report on November 18 claiming its AI models have delivered measurable improvements in both its custom inference chip, Jalapeño, and AMD's flagship GPU and CPU designs. The company says the work compressed roughly a year of engineering progress into less than five months, with some changes already incorporated into designs sent for manufacturing.

Jalapeño: Area Reductions Verified in Silicon Bring-Up

OpenAI says it began using an early model, GPT‑o3 mini, on Jalapeño in early 2025. The model proposed changes to the chip's vector and matrix compute blocks—redesigning floating-point adders, multipliers, conversion logic, and pipeline organization. The company estimates the model-driven changes reduced standard-cell area by about 13% in the vector unit and 10% in the matrix unit. One multiplier rewrite cut cell area by roughly 15% while reducing latency from two clock cycles to one; an adder redesign similarly shaved about 15% area while keeping its two-cycle latency. OpenAI says these optimizations were "silicon proven during Jalapeño's silicon bring-up," the post-manufacturing validation phase [1].

Jalapeño is OpenAI's first custom inference accelerator, developed with Broadcom and Celestica. The company previously said it went from design to tape-out in nine months, with AI models accelerating parts of the process [2].

AMD Collaboration: Up to 14% PPA Gain on Flagship GPU

OpenAI reports a broader collaboration with AMD on power, performance, and area (PPA) improvements. On AMD's flagship GPU intellectual property—already described as "heavily optimized"—OpenAI's models rewrote SystemVerilog RTL, the code that describes circuit logic, and identified further gains of up to 14% relative to that baseline. The company says this represents roughly a year of progress achieved in less than 5 months [1].

AMD has incorporated some of the AI-generated changes into a recent design sent for fabrication (tape-out) and added others to production design files for future versions. AMD also verified single-digit percentage PPA improvements in key blocks of its upcoming next-generation flagship CPU using designs written by OpenAI's model [1]. OpenAI President Greg Brockman separately posted on X that the results represent "roughly an additional year of progress on @AMD's chip achieved in less than 5 months using our models, with some of these results included in a design that's recently been taped out" [4].

Benchmarking Model Generations on RISC-V Cores

To track progress across model generations, OpenAI tested five models—GPT‑5.5, GPT‑5.5 Sol, GPT‑6.1 Sol, GPT‑6 Astra, and GPT‑6.2 Astra—on two open-source 32-bit RISC-V cores: PicoRV32 and Ibex. The tasks were (1) optimizing an existing RTL description for throughput, and (2) designing a core from a written specification. At a comparable compute budget of about 112,000 tokens per run, GPT‑6.2 Astra scored highest on both cores for both tasks [1].

OpenAI acknowledges limitations in its benchmark methodology. When designing from a specification, models could omit features present in the original cores—such as performance counters and extra instructions—that may not improve throughput on the test workloads but would be required in production. The company notes that "requiring those features, setting a tighter limit on circuit area, and checking timing after physical synthesis would make the test more representative of production chip design" [1]. Designs that failed simulation or exceeded 16× the original area were scored zero. The full evaluation comprised 9,881 scored runs.

Synopsys Partnership Aims to Embed Models in EDA Workflows

OpenAI announced a multi-year partnership with Synopsys on September 30, 2026, to develop GPT‑Synopsys—a specialized model trained to operate Synopsys electronic design automation (EDA) tools as expert engineers do [7]. The goal is to move beyond code suggestion to having agents run tools, interpret results, modify circuits, and iterate toward verified outcomes for engineer review. OpenAI says early technology engagements are underway with leading semiconductor customers [1][7].

What Is Verified vs. What Is Claimed

  • Jalapeño improvements: OpenAI says the area reductions were validated during silicon bring-up—physical testing of manufactured chips. This is the strongest form of verification cited.
  • AMD GPU gains: The up-to-14% PPA figure is relative to AMD's own heavily optimized baseline, as measured by OpenAI's models using Synopsys tools. AMD has taken some changes to tape-out and others into future design files. Independent third-party verification of the 14% figure is not cited.
  • AMD CPU gains: Described as "single-digit percentage" on key blocks, verified by AMD. No further detail on which blocks or how measured.
  • Timeline claim: OpenAI characterizes the GPU work as roughly a year of progress achieved in less than 5 months, a comparison echoed by Brockman. It reflects the company's internal comparison, not an industry-standard metric.
  • RISC-V benchmarks: Useful for tracking model-to-model progress, but OpenAI itself flags that the specification-based task allows omissions that inflate scores relative to production requirements.

Context

The announcement extends OpenAI's push into the hardware layer powering its models. With inference costs rising, custom silicon and faster design cycles are strategic levers. The Synopsys partnership signals an intent to productize AI-assisted chip design as a service, not just use it internally. Whether the reported PPA gains translate to shipping products with measurable performance-per-watt advantages will become clear as AMD's next-generation GPU and CPU reach market.

Sources

  1. Advancing chip design with OpenAI models (openai.com)
  2. AI chip design: OpenAI says its models improved AMD GPU IP (yourstory.com)
  3. AI TRENDS | OpenAI Says Its Models Accelerated AMD GPU Design and ... (binance.com)
  4. 'Unprecedented Results': AMD Packs a Year of Chip Progress Into 5 Months ... (finance.yahoo.com)
  5. OpenAI claims its models have delivered chip design ... (datacenterdynamics.com)
  6. Techmeme (techmeme.com)
  7. Synopsys, Inc. | OpenAI and Synopsys Announce GPT-Synopsys: Frontier Intelligence to Revolutionize Chip Design (investor.synopsys.com)