AI/ ai · chip design · openai · synopsys

OpenAI and Synopsys Build an AI Model to Operate Chip Tools

A new OpenAI-trained model will run Synopsys's chip design tools on its own, deepening an AI arms race already drawing in Cadence and Siemens EDA.

OpenAI and Synopsys are building an AI model that can operate chip-design software on its own.

The two companies signed a multi-year agreement on September 30 to jointly develop GPT-Synopsys, a model tuned specifically to run Synopsys's electronic design automation (EDA) tools. Engineers would hand the model goals like timing closure or power-performance-area optimization, and agents would run the tools, read the results, make changes, and iterate until the design is ready for a human to review. OpenAI is licensing Synopsys's software to train the model, which will run on OpenAI's own infrastructure, and the two companies plan to split the revenue once it ships. Neither a release date nor pricing has been announced.

This matters because every major EDA vendor is now racing to put agents in charge of chip design, not just assist with it. Synopsys has been layering AI onto its tools since 2020's DSO.ai, and rival Cadence already sells ChipStack, an agent system built on Nvidia and Google models. Siemens EDA is pitching its own autonomous-design agents too, which means GPT-Synopsys's real pitch isn't "AI helps engineers" - it's "a model trained specifically on one company's toolchain can out-reason a general-purpose one bolted onto the same software." OpenAI has its own stake here too: it used AI heavily to help design the Jalapeno inference chip with Broadcom, taking it from concept to tapeout in nine months.

Whether any of this saves money is unclear. Neither company has published pricing, and other AI deployments - including, by some accounts, at Uber - have shown that handing a process to AI agents can cost more than paying engineers to do it, not less.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →