# Advancing chip design with OpenAI models

**URL:** <https://community.openai.com/t/advancing-chip-design-with-openai-models/1417964>\
**Category:** Community\
**Tags:** announcement, research, openai\
**Created:** [November 18, 2026, 7:03pm UTC](https://community.openai.com/t/advancing-chip-design-with-openai-models/1417964 "2026-11-18T19:03:10Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![DysTopia](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/dystopia/32/711735_2.png) [@DysTopia](https://community.openai.com/u/DysTopia)\
**Post date:** [November 18, 2026, 7:03pm UTC](https://community.openai.com/t/advancing-chip-design-with-openai-models/1417964/1 "2026-11-18T19:03:10Z")

</div>

# Advancing chip design with OpenAI models

AI helping design the chips that power AI… that’s a neat little feedback loop. OpenAI’s latest update shares how its models are helping engineers improve chip designs, with gains already demonstrated in its own Jalapeño chip and further improvements heading into AMD hardware. An interesting peek at how AI is helping shape the hardware behind it. [OpenAI](https://openai.com/index/advancing-chip-design-with-our-models/?utm_source=chatgpt.com)

 ![PicoRV32 · RTL optimization](https://us1.discourse-cdn.com/openai1/original/4X/0/5/0/050f18302e2689933c554555e9bf2a7c1b3c7271.png)  
 ![Ibex · RTL optimization](https://us1.discourse-cdn.com/openai1/original/4X/7/f/9/7f95b003c31970b9cbfde45ed820d2bda2b717ff.png)  
 ![PicoRV32 · Specification to RTL](https://us1.discourse-cdn.com/openai1/original/4X/a/f/b/afb728ee81e2e37d2f27c43c71ac96367c681952.png)  
 ![Ibex · Specification to RTL](https://us1.discourse-cdn.com/openai1/original/4X/0/c/6/0c6f65ffda6adab606adf922e4111d5f99224b0c.png)

Curious to see what engineers build with these tools next. With a chip called Jalapeño, I suppose things were bound to get a little spicy. 🌶😏

Enjoy!

Dys

---

<div class="post-metadata">

**Author:** ![Diet](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/diet/32/562749_2.png) [@Diet](https://community.openai.com/u/Diet)\
**Post date:** [November 18, 2026, 7:18pm UTC](https://community.openai.com/t/advancing-chip-design-with-openai-models/1417964/2 "2026-11-18T19:18:08Z")

</div>

I think it’s an excellent demonstrations of using LLMs to empirically explore the configuration space of whatever problem’s at hand.

It’s not that these improvements are amazing on their own, IMO, but rather that we can do _more_ engineering work per engineer. _Super good enough_ becomes _eh, let it crank?_

The challenge for humans now is to set up these problems.

Frontline engineers have graduated from _doing_ the homework to _creating_ homework.

This is a beautiful example of this.

---

<div class="post-metadata">

**Author:** ![platypus](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/platypus/32/759972_2.png) [@platypus](https://community.openai.com/u/platypus)\
**Post date:** [November 18, 2026, 7:22pm UTC](https://community.openai.com/t/advancing-chip-design-with-openai-models/1417964/3 "2026-11-18T19:22:05Z")

</div>

This is awesome. I’ve maintained that the next frontier is the physical frontier - using AI to help create hardware, robotics, mechatronics, learn about electronics and physical builds.

Then there is the additional aspect here of AI labs like OpenAI using their own models to build self-sufficiency across the full stack, like the chips for running their own models on. Great stuff!

---

<div class="post-metadata">

**Author:** ![martinberger](https://avatars.discourse-cdn.com/v4/letter/m/f17d59/32.png) [@martinberger](https://community.openai.com/u/martinberger)\
**Post date:** [November 19, 2026, 5:18pm UTC](https://community.openai.com/t/advancing-chip-design-with-openai-models/1417964/4 "2026-11-19T17:18:22Z")

</div>

Hello, this was a fascinating experiment. I’m particularly interested in the experiments generating PicoRV32 and Ibex RTL from specifications. I could not quite work out from the article how the RISC-V core was specified. I am not asking about the prompt, but did the specifications included the official executable RISC-V Sail ISA model or whether they consisted primarily of natural-language requirements and RTL interface templates?

Also, were generated processors checked against a formal ISA reference model, or exclusively through simulation-based tests? The official RISC-V specification in Sail automatically generates a suitable simulator and a logical specifications (e.g. in SMT-LIB) that would be used here.
