OpenAI says its own AI models are now designing better chips. In a post published on 18 November, the company said model-proposed changes cut standard-cell area by about 13% in the vector unit and 10% in the matrix unit of Jalapeño, its first custom inference chip — and that those optimisations were proven in working silicon during the chip’s bring-up.
OpenAI started using GPT-o3 mini to help design Jalapeño in early 2025. In the vector unit, which handles parallel computing tasks, the models found savings in floating-point adders and multipliers, conversion functions and reduction logic. In the matrix unit, one redesigned multiplier used about 15% less cell area while cutting latency from two clock cycles to one, and a redesigned adder cut area by about 15% while keeping its two-cycle latency.
The same approach is now being applied to other companies’ hardware. OpenAI said it partnered with AMD on power, performance and area improvements and achieved gains of up to 14% on flagship GPU designs compared with heavily optimised baselines — roughly a year of progress in under five months. AMD has already included some of the changes in a design it sent for manufacturing and added others to its production design files. AMD also verified single-digit percentage gains on key blocks of its next-generation flagship CPU.
To measure how the capability is improving, OpenAI tested successive model generations on two open-source 32-bit RISC-V processor cores, PicoRV32 and Ibex. With a comparable budget of about 112,000 tokens per run, its newest model, GPT-6.2 Astra, had the highest average score on both cores. The full evaluation covered 9,881 scored runs across both tasks, both cores and all token budgets.
OpenAI is also working with Synopsys to train GPT-Synopsys, a model designed to use the electronic design automation tools that chip engineers already use to design and verify chips.
What it means
AI is no longer just running on chips — it is helping design them, and the gains are showing up in production hardware, not just lab demos. If models keep improving at this pace, chip design cycles could shorten sharply, and the companies that own the best AI models may also gain an edge in the hardware those models run on.
Sources:
- OpenAI, “Advancing chip design with OpenAI models”, 18 November 2026 — https://openai.com/index/advancing-chip-design-with-our-models/
- ExplainX, “OpenAI AMD Chip Design: 14% PPA in 5 Months (Nov 2026)” — https://explainx.ai/blog/openai-amd-jalapeno-chip-design-astra-ppa-november-2026

