The model is meant to go beyond a general-purpose assistant bolted onto software. Synopsys's EDA engines would handle the calculations and checks, while GPT-Synopsys interprets the outputs, makes engineering judgements, and iterates towards verified outcomes for a human to review. Engineers would hand it objectives such as power, performance and area optimisation, timing and verification closure.

There is history here. Synopsys launched its DSO.ai reinforcement learning tool in March 2020, added generative AI assistance through a Microsoft collaboration in November 2023, and unveiled its agentic Autopilot platform and AgentEngineer portfolio on September 28, two days before the OpenAI deal. OpenAI has also seen the upside first hand: it says its Jalapeño inference accelerator, built with Broadcom, went from initial design to tape-out in nine months with heavy help from its own models. The deal extends a loop the industry has been circling for years: as of late August, Architect Labs claimed an industry first with a chip designed almost entirely by AI, and today's models increasingly help design the processors their successors will run on.

Open questions remain. Hosting on OpenAI infrastructure raises the usual worries about confidential designs; the companies say customer data will be excluded from training and encrypted, but the announcement names no hosting regions or default retention periods, and offers no contractual terms on ownership of generated outputs or third-party licensed design IP.