Synopsys and OpenAI are developing a specialized artificial-intelligence model intended to operate chip-design software, a potentially important shift from AI that advises engineers to AI that can execute portions of an electronic-design workflow.

The companies announced the multi-year agreement Wednesday, saying GPT-Synopsys will combine OpenAI models with Synopsys electronic design automation, or EDA, tools. Under the agreement, OpenAI will license Synopsys tools while the companies jointly conduct research and sell the resulting service. Reuters reported that OpenAI will pay a training subscription fee and that the partners will share revenue when customers use the product.

The significance lies less in a conversational interface than in tool use. Chip development moves from circuit descriptions through synthesis, physical layout, timing analysis and multiple verification stages. Each stage forces tradeoffs among performance, power consumption and silicon area. The companies say GPT-Synopsys will be trained to run EDA tools, interpret their outputs, revise a design and repeat the process until it reaches an engineer-defined objective.

From assistant to operator

Synopsys already markets AI-assisted and agentic design products. On Monday, it introduced an Autopilot platform and seven AgentEngineer systems spanning verification, implementation, analog design, manufacturing and simulation. The company said more than 50 customer engagements were underway and general availability was planned for the end of 2026. An independent industry review placed that rollout in a broader competition among Synopsys, Cadence, Siemens and other EDA vendors to automate longer engineering tasks.

GPT-Synopsys is designed to sit above and alongside those tools. The joint announcement describes engineers delegating goals such as improving power, performance and area or closing timing and verification gaps. Agents would then run tools, read results and make changes for an engineer to review. The model is expected to operate on OpenAI-hosted infrastructure and integrate with Synopsys.ai and Autopilot, while also working with customers’ own agent-management systems.

That approach could let engineering teams explore more alternatives than they can test manually. OpenAI President Greg Brockman said the aim is to cut weeks or months from design schedules. But the companies have not disclosed a release date, pricing, benchmarks or named customers for the new service. Early technical engagements are underway, according to their announcement, which means the productivity claims remain prospective rather than independently demonstrated.

Verification remains the boundary

The partners are not proposing that a language model become the final authority on whether a chip works. Synopsys Chief Executive Sassine Ghazi told Reuters that GPT-Synopsys output will still be checked by conventional software using deterministic methods. In semiconductor development, those sign-off systems test whether a design satisfies timing, electrical, thermal and manufacturing rules before a company commits to fabrication.

That separation is central to the deal’s credibility. Generative models can propose changes and navigate complex software, but they can also produce incorrect or inconsistent results. A flawed answer in ordinary office work may be repairable; an undetected flaw in a chip design can become extremely expensive once masks are made and wafers enter production. GPT-Synopsys may accelerate the search for a workable design, while established verification tools remain responsible for grounding the result in physics.

Security is another unresolved test because unreleased chip designs are among a semiconductor company’s most sensitive intellectual property. Synopsys and OpenAI say customer design data will not be used to train the model, will be encrypted in transit and at rest, and will be subject to configurable retention, audit and permission controls. Those are company commitments; customers will still need to evaluate architecture, access controls, data residency and incident-response terms before placing proprietary designs in a hosted workflow.

A strategic bet on AI-driven complexity

The partnership also gives Synopsys a way to monetize AI beyond ordinary software subscriptions. Ghazi said revenue sharing will depend on the value the model adds to a customer’s design, although the parties did not disclose the formula. Synopsys presented the agreement during an investor meeting at which it forecast 15% revenue growth in fiscal 2027, above the 11.19% analyst consensus cited by Reuters.

The deal builds on a strong period for the company’s core design business. Synopsys reported $2.48 billion in fiscal third-quarter revenue, up 42% from a year earlier, with much of the increase reflecting a full quarter of Ansys ownership. Its quarterly filing also said EDA demand was broad-based and highlighted regulatory, export-control and integration risks that could affect future performance.

For chipmakers, the practical question is not whether AI can generate a plausible layout suggestion. It is whether an agent can reliably operate a long chain of specialized tools, preserve confidential design intent and improve measurable outcomes without weakening verification. GPT-Synopsys is a serious attempt to answer that question, backed by one of the dominant EDA suppliers and a major model developer. For now, however, it is a development program with commercial terms—not yet proof that autonomous chip design has arrived.