GPT-Synopsys is the joint project OpenAI and Synopsys announced on September 30: a specialized model tuned for semiconductor design workflows, built to teach AI agents to operate EDA tools like seasoned engineers.
According to Synopsys' official announcement, the two companies signed a multi-year strategic partnership to co-develop GPT-Synopsys. The agreement includes a revenue-sharing arrangement and joint go-to-market plans, with OpenAI licensing Synopsys' electronic design automation (EDA) tools to train the model. Synopsys president and CEO Sassine Ghazi and OpenAI co-founder Greg Brockman both publicly backed the deal, and Reuters followed up with its own reporting.
How the agents will work: run tools, read results, optimize again
Today's agent technology can already connect general-purpose models to EDA tools to run chip-design workflows. GPT-Synopsys aims to go one step further: turning frontier models into "native expert users" of EDA tools.
The workflow looks like this: after engineers set targets for power, performance and area (PPA), agents call Synopsys tools to run the design, interpret the results, revise the design accordingly, and repeat the loop, with engineers reviewing the output at the end. Brockman said the process could "shave off weeks, months from the design process" and bring more chips to the world.
A few details in the agreement stand out. The model will run on OpenAI-hosted infrastructure, integrate with Synopsys.ai and Synopsys Autopilot, and interoperate with customers' own agent-harness systems; early technology engagements are already underway with semiconductor customers. On data, Synopsys pledged that customer-proprietary design data will not be used to train the model, with encryption in transit and at rest plus configurable retention, audit and permission controls.
Ghazi drew a red line in his Reuters interview: the model's output will still be double-checked by Synopsys' traditional sign-off tools using conventional computing techniques. In his words, the model needs these "guardrails" to "check the physics" — a chip that tapes out wrong costs real money.
Another signal from the same investor summit
Synopsys chose its investor summit to announce the deal, and used the occasion to raise its fiscal 2027 revenue growth outlook to 15%, above the analyst consensus of 11.19%. Shares jumped more than 7% on the news.
This is not OpenAI's first move into the chip supply chain. It has already partnered with Broadcom on custom silicon for AI workloads, including the recently disclosed Jalapeno chip. This deal extends that reach upstream into design: from "building its own compute" to "helping others design chips."
Who feels the change
For chip engineers, the workflow may shift from "tuning parameters by hand" to "setting targets and reviewing results." Models can explore more design alternatives and converge on PPA faster, leaving engineers to focus on trade-off judgment rather than repetitive tuning.
For the EDA industry, the business model is loosening. Synopsys used to sell tool licenses; under this agreement, the two sides share revenue based on how much the model actually improves a design — the tool vendor starts selling the effectiveness of an "agent workflow," not just software. Ghazi made a point of saying the deal "will not be cannibalizing our business," which suggests the industry is not without misgivings about that shift.
For the AI industry, this is another step of frontier models into hardcore engineering verticals. After code, documents and customer service, agents are taking on professional workflows with physical constraints and no room for error. But the boundary is worth seeing clearly: this is a development agreement, not a finished product for sale. The real test will be how much time GPT-Synopsys actually saves in a real tape-out flow.
