OpenAI's Models Redesigned AMD's Flagship GPU — 14% Better in Under Five Months

Photo: Hippopx (public domain). Illustrative image.
What happened. OpenAI published a technical report on 18 November 2026, “Advancing chip design with OpenAI models”, showing its frontier models doing chip design work that makes it into real silicon. Working with AMD, OpenAI says its models rewrote parts of the SystemVerilog RTL — the code that describes a chip’s circuits — for AMD’s flagship GPU IP, and used Synopsys design software to verify each rewrite still worked and to measure the trade-offs. The reported result: up to 14% improvement in PPA, the combined measure of power, performance and area, against baselines that were already heavily optimised. OpenAI calls that roughly a year of design progress in less than five months.
The changes are already in hardware. According to OpenAI, AMD has included some of the model-proposed changes in a recent design it sent for manufacturing, and added the rest to its production design files for future versions. AMD also verified single-digit percentage PPA gains on key blocks of its next flagship CPU using model-written designs. OpenAI names no AMD product, gives no breakdown of the 14% between power, performance and area, and reports no measurements from finished AMD silicon — so the figure does not mean the GPU runs 14% faster. AMD has published no statement of its own on the work as of 19 November.
Jalapeño is where the loop closes. The AMD work grew out of OpenAI’s own inference chip, Jalapeño, unveiled with Broadcom as OpenAI’s first Intelligence Processor. Engineers began using GPT-o3 mini to help design Jalapeño in early 2025, and the chip went from design to tape-out in nine months. OpenAI says model-proposed changes cut standard-cell area by about 13% in the vector unit and 10% in the matrix unit, confirmed when the actual chip was brought up in the lab — including one multiplier redesign that used about 15% less cell area while cutting latency from two clock cycles to one. Engineering samples are running ML workloads at target frequency and power, including GPT-5.3-Codex-Spark, with initial deployment planned by the end of 2026 and gigawatt-scale rollout with data-centre partners from 2026.
The test behind the claims. OpenAI ran 9,881 scored evaluations across two open-source 32-bit RISC-V cores, PicoRV32 and Ibex, asking models either to optimise existing designs or build them from written specifications. GPT-6.2 Astra scored highest on both under a roughly 112,000-token budget per run. OpenAI is candid about a caveat: models scored higher designing a core from a specification than improving the real one, partly because they could leave out features such as performance counters that the benchmark did not reward. OpenAI is now working with Synopsys to train GPT-Synopsys, a model that uses professional electronic design automation tools the way an expert chip designer would.
Why it matters. This is the first public case of AI models improving the flagship silicon of a major chipmaker, not just a lab demo — and of the same models improving the chips that will run them. If the gains hold up in finished silicon, AI-designed hardware becomes a compounding advantage: better models design better chips, which run better models. The next checks are an AMD statement naming the products that carry the changes, and finished-silicon measurements. For the hardware race behind all of this, see Google’s MLPerf Ironwood results.
Sources
- OpenAI — “Advancing chip design with OpenAI models” (18 November 2026): https://openai.com/index/advancing-chip-design-with-our-models/
- OpenAI — “OpenAI and Broadcom unveil LLM-optimized inference chip” (Jalapeño): https://openai.com/index/openai-broadcom-jalapeno-inference-chip/
- Synopsys — “OpenAI and Synopsys Announce GPT-Synopsys” (30 September 2026)
- FreshSynapse — “GPT-6.2 Astra and Sol Are Here: OpenAI’s Frontier Model Gets 20% Cheaper”: https://freshsynapse.com/articles/gpt-6-2-astra-sol-launch-2026-11-18/
- FreshSynapse — “Google’s Ironwood TPUs Post 10 MLPerf v6.1 Training Results”: https://freshsynapse.com/articles/google-mlperf-v6-1-ironwood-tpu-results-2026-11-18/