OpenAI and Synopsys Plan AI Model to Help Design Chips

OpenAI and Synopsys are working on an AI model that will help engineers design chips. That sounds like a natural pairing until you consider how exacting the job is. A chip design must satisfy thousands of constraints before it can be manufactured, and a confident-looking answer from a chatbot is not enough. The companies say their planned system, called GPT-Synopsys, will operate established design tools and put its results in front of engineers for review.

Under a multi-year agreement announced September 30, OpenAI will license Synopsys electronic design automation software and use it to develop a specialised model for semiconductor workflows. The partners also plan joint research and a commercial offering with shared revenue. Early technology engagements with chip customers are already under way, but neither company has announced a general release date or measured time savings for the finished product.

The proposed difference from an ordinary AI assistant is that GPT-Synopsys would work inside the tools engineers already use. A design team could give it an objective, such as improving power efficiency or resolving a timing problem. The model would run the relevant software, interpret the output, make changes and test again. An engineer would still have to review the result. In chipmaking, even a small error can become very expensive once a design reaches manufacturing.

Synopsys describes this work in terms of power, performance and area, usually shortened to PPA. Those demands pull against each other. Making a processor faster, for example, can increase heat or take up more space on the silicon. The attraction of an AI agent is not that it magically finds one perfect answer. It may be able to explore more trade-offs than a team has time to examine by hand, while using trusted tools to check each attempt.

There is a circular quality to the deal. AI companies need better chips to train and serve models; now one of those companies wants its models to help design the chips. OpenAI has also been building out its own compute strategy, including a broader search for specialised AI hardware. Working with Synopsys gives it a route into the engineering software behind many chip projects, not just the hardware it buys for itself.

The companies say GPT-Synopsys will run on OpenAI-hosted infrastructure and work with customer agent systems. They also promise that customer design data will not be used to train the model, with encryption and controls for retention, permissions and audits. Those assurances matter because chip layouts and verification work can be among a company’s most valuable trade secrets. Customers will want to test the safeguards as closely as they test the model’s engineering claims.

For now, this is a development agreement rather than proof that a chip has been designed faster or better. The useful question is what happens when the system faces a real design with competing requirements, incomplete information and a hard deadline. If it can make that work more efficient without weakening verification, the benefit could spread well beyond AI accelerators to the many devices that depend on custom silicon.