OpenAI and Synopsys Team Up on GPT-Synopsys for Chip Design
OpenAI and Synopsys signed a multi-year deal to build GPT-Synopsys, a model trained to operate Synopsys' chip-design tools itself. OpenAI will license the EDA tools, and the two will share revenue.
TL;DR
- OpenAI and Synopsys announced a multi-year partnership on September 30 to develop GPT-Synopsys, a specialized model optimized to run Synopsys' chip-design software [^1].
- OpenAI will license Synopsys' electronic design automation (EDA) tools to build the model, and the companies will share revenue from selling it [^1].
Background
Chips are designed with electronic design automation software, and Synopsys is one of the leading vendors. The press release says today's agentic AI connects general-purpose models to such tools to run design workflows, and the stated next step is to make frontier models expert users of the tools themselves: running them, interpreting their output and iterating on designs [^1].
What happened
Under the agreement, the two companies will jointly develop GPT-Synopsys and collaborate on research and go-to-market under a shared revenue framework [^1]. According to the announcement, engineers will be able to delegate design objectives, from optimizing power, performance and area to closing timing and verification, to agents that run the tools, interpret results, implement changes and iterate toward verified results for engineer review [^1]. The model will run on OpenAI-hosted infrastructure, interoperate with customers' own agent harnesses and integrate with Synopsys.ai and its Autopilot agentic platform [^1].
The service will bundle compute, model and licenses. Synopsys says customer design data will not be used for training, will be encrypted at rest and in transit, and can be managed through retention, audit and permission controls [^1]. Early engagements with leading semiconductor customers are underway [^1]. "We're using our most advanced technology to improve the systems that power AI," said OpenAI president Greg Brockman [^1].
Why it matters
Chip design is a natural target for agents: the tools are complex, each run is expensive, and an engineer's time goes into setting up runs and reading results. A model trained to use the tools natively, rather than through a generic connector, could explore more design options in the same time. For OpenAI there is also a loop: better chips for AI, designed partly by AI. The announcement makes promises rather than reporting results, so how much faster designs actually get remains to be seen.
Practical example
Say a chip team misses its timing target on one block. Today an engineer adjusts settings, reruns the tool overnight and reads the reports in the morning. In the workflow Synopsys describes, the engineer would hand the timing-closure goal to the agent, which runs the tools, reads the results and iterates, then presents a verified result for the engineer to review [^1].