A frontier AI lab will pay to learn how to design chips, and the two sides will split the proceeds when the results reach customers.
That is the shape of the multi-year agreement between Synopsys and OpenAI. The model, GPT-Synopsys, is trained to operate electronic design automation software: running the tools, reading their output and iterating toward a better design, the same tasks an engineer performs by hand.
Money flows both ways. OpenAI licenses the EDA tools and pays for the training phase, and later revenue is divided according to how much the model improves a given design. Compute, model access and tool licenses are sold as a single package.
Both executives framed the work as widening access. Synopsys boss Sassine Ghazi said more companies would be able to reach advanced design, while OpenAI’s Greg Brockman tied shorter design cycles to putting more silicon into circulation.
Verification stays conventional. Ghazi said each result is re-checked with ordinary computational methods to confirm the chip will function, calling that physical proof essential rather than optional.
The pair say customer design data is encrypted and access-controlled, is never used for training, and that early technical work is running with unnamed leading customers.
