GPT-Synopsys for chip design will combine OpenAI models with Synopsys electronic-design-automation tools under a multi-year partnership announced September 30. The companies plan to train the specialized model to reason about semiconductor designs, operate engineering software and iterate toward results that human engineers can review.

 

The agreement goes beyond adding a chatbot to an existing product. OpenAI will license Synopsys tools for development, the companies will jointly sell a bundled service, and their commercial framework includes both a training subscription and shared revenue.

 

The agreement establishes five components:

  • OpenAI licenses Synopsys EDA tools
  • The model targets design and verification work
  • The service bundles compute, models and software licenses
  • OpenAI and Synopsys share customer revenue
  • Customer design data is excluded from training

 

OpenAI and Synopsys Launch GPT-Synopsys

The companies described GPT-Synopsys as a model optimized specifically for semiconductor workflows. According to the official announcement, it will connect OpenAI's model technology with Synopsys' EDA software and engineering knowledge rather than rely only on general-purpose reasoning.

 

Synopsys said early technology engagements are already underway with leading semiconductor customers, but it did not name them. The partners intend to make the service available globally and will collaborate on research, development and sales.

 

The announcement does not include a public release date, pricing or measured performance results. That distinction matters because a development agreement and early customer testing do not yet establish how much time or money the system can save in production.

 

How GPT-Synopsys Will Operate EDA Tools

Chip design begins with code-like descriptions of circuits and proceeds through logic, timing, power, verification and physical layout. Each stage creates trade-offs among performance, power consumption and silicon area, while errors discovered late can delay manufacturing and sharply increase costs.

 

GPT-Synopsys is intended to work inside that iterative process. Engineers could assign objectives such as improving power, performance and area or reaching timing and verification closure. Agents would run Synopsys tools, interpret their outputs, change the design and repeat the cycle before presenting a verified result for review.

 

This approach differs from generating a block of hardware code once. The model must respond to tool output and maintain progress across many steps, using established engineering software as both an execution environment and a source of feedback.

 

Traditional verification remains essential. Synopsys chief executive Sassine Ghazi told Reuters that the model's work will still be checked by conventional tools that validate whether a proposed chip satisfies physical and sign-off requirements.

 

Related Research

 

GPT-Synopsys for Chip Design Uses Shared Revenue

The financial structure is unusual for enterprise software. Reuters reported that OpenAI will pay Synopsys a training subscription fee while learning to use its tools. When customers use the finished product, the companies will share revenue based on the model's contribution to improving a chip design.

 

That structure links compensation to customer value more directly than a simple software-seat license. It could also protect Synopsys from having an AI layer replace its core products, because the model depends on those products to execute workflows and validate results.

 

Neither company disclosed the subscription amount, revenue split or method for measuring improvement. Those details will determine whether the arrangement can scale across customers whose designs, tool configurations and engineering targets vary widely.

 

Synopsys separately told investors that it is pursuing subscription and consumption-based pricing for AI services. GPT-Synopsys therefore serves as both a technical project and a test of whether specialized models can create a new recurring-revenue layer around established engineering platforms.

 

OpenAI Hosting Raises Security and Product Questions

GPT-Synopsys will run on OpenAI-hosted infrastructure and is designed to work with customer agent harnesses. Synopsys also plans deep integration with Synopsys.ai and its Autopilot platform, giving chip companies multiple routes to connect the model with existing engineering systems.

 

The partners say customer-specific design data will not be used to train the model. They also promise encryption at rest and in transit, configurable retention, audit controls and permission management, all important requirements for semiconductor companies protecting valuable intellectual property.

 

Those commitments still leave deployment questions. Customers will need to evaluate data residency, access boundaries, logging, model isolation and the handling of proprietary process information before allowing an external agent to operate inside sensitive design flows.

 

The strongest evidence will come from named deployments and reproducible measures: fewer engineering iterations, faster timing closure, better power-performance-area results or shorter paths from specification to sign-off. Until those results appear, GPT-Synopsys is a strategically important partnership with a defined architecture, not a proven productivity benchmark.

 

If it works, the system could establish a template for vertical AI in high-value technical fields. The model would not replace the authoritative tools or the engineer; it would learn to operate the tools repeatedly, while verified software and human review remain the final controls.