OpenAI is paying Synopsys for the privilege of learning to use its software. That is the arrangement at the centre of a multi-year deal the two announced on Wednesday to build GPT-Synopsys, a model trained to reason about chip design and to operate Synopsys' tools directly.
OpenAI licenses the electronic design automation software — the programs used to design most of the world's most advanced semiconductors — and, according to Synopsys chief executive Sassine Ghazi, pays a training subscription fee for the lessons. When customers eventually use the product, the two companies split the revenue based on how much the model actually improves the chip.
Run that against the expectation everyone has been carrying for three years. A model ingests a professional domain, learns what the specialists know, and the software those specialists used becomes a thin wrapper nobody pays much for. That is the story of this era in a sentence, and it is the reason specialist software companies trade on nerves. Here the model arrives as a licensee. It pays to be taught, and then it earns in proportion to how useful it turns out to be, which is not how software licensing usually works and is very much how a consultant gets paid.
Ghazi did not leave the point implicit. "We structured the agreement in a way that it will not be cannibalizing our business," he told Reuters, and the structure shows where he drew the line. The model handles the trade-offs and optimisations — the judgement-heavy middle of chip design, which Greg Brockman says could come down by weeks or months. But Ghazi says the verification stays where it was: Synopsys' conventional, non-AI tools still determine whether a proposed design actually works. The model does the trade-offs; the old software still decides whether the chip works.
What neither company published is everything that would let anyone check the leverage. The revenue split is undisclosed. So are the exclusivity terms. The press release says customer design data is protected and will not be used to train the model, which answers the question customers would ask first and leaves the commercial one open. And Synopsys is hardly holding out against the AI companies: Nvidia put $2 billion into it last December as part of an expanded tie-up to build design tools together. This is the third such arrangement, not a stand.
The market priced it as the good kind. Shares rose as much as 7 percent, and at the same investor summit Ghazi lifted the fiscal 2027 revenue growth outlook to 15 percent, against an analyst consensus of 11.19 percent. Whatever else the deal is, the company that was supposed to be disrupted raised its forecast on the day it signed.
Everyone assumed the model would replace the software. The software charged it tuition, kept the final say, and raised its guidance.
The model does the trade-offs; the old software still decides whether the chip works.
