OpenAI and Synopsys announced a multiyear partnership on September 30 to build GPT-Synopsys, a specialized AI model that operates Synopsys' chip design software. The model is meant to run electronic design automation tools, read their results, revise a design and verify it again. The companies plan to sell it as a service and split the revenue. There is no release date, no pricing and no published benchmark yet, so the headline claim of faster chip design is still unproven.

Table of contents

What GPT-Synopsys is

Synopsys makes electronic design automation software, the tools engineers use to design and verify chips. Per a summary of its release, the new model combines OpenAI's frontier models with that software so it can reason about chip design and verification and operate the tools directly. Business Standard describes the loop: use the tools, interpret what comes out, change the design, repeat.

The model will run on OpenAI-hosted infrastructure, according to HPCwire's coverage of the announcement. It is meant to work with customers' own agent systems and to plug into Synopsys.ai and Synopsys Autopilot, the agentic AI platform Synopsys announced two days earlier, on September 28, as Xenospectrum notes. R&D World reports both companies say customer data will not be used to train the model.

The news came during Synopsys' investor day on Wednesday. CEO Sassine Ghazi said the model should widen access to Synopsys' design tools, per The Stack.

How the deal makes money

The structure is unusual. Ghazi said in an interview, as summarized by BigGo Finance, that OpenAI will pay Synopsys a training subscription fee while the model learns the tools. OpenAI licenses the EDA software for that purpose, according to Business Standard. Once customers use the product, the two companies split revenue based on how well it improves a chip's design.

That last part is the interesting one. A revenue split tied to results, not usage, puts both companies on the hook for the model working. It also means Synopsys keeps earning as chip production grows. R&D World points out that the announcement discloses no royalty rate or payment schedule, so the actual split is unknown.

Synopsys announced a separate agreement with Amazon the same day, covering reusable circuit design assets, per R&D World. That deal is about custom silicon and does not name Amazon as a GPT-Synopsys customer.

What it does not claim

This is not an announcement that AI can now design a finished chip and send it to a factory without people. Xenospectrum states that plainly, and the company's own comments back it up. Ghazi said the model's output will still be checked by Synopsys tools that use conventional computing to confirm a chip will work, according to BigGo Finance.

That check matters. A chip is expensive to fix after manufacturing, and a language model can produce plausible but wrong designs. Using traditional verification as a backstop is the sensible design, though it also means the model's value depends on how many bad candidates it wastes verification time on. Nobody has published that number.

Why now

The Stack notes OpenAI recently launched its first custom AI chip with Broadcom, so it has its own reasons to want better design tools. Chip makers face rising complexity at advanced nodes, and PromptZone says teams there spend thousands of engineer-hours on verification and physical implementation. If a model can shorten even part of that, the savings are large.

The demand for custom chips is also tied to the compute race we have been following. We covered the scale of commitments in our report on Anthropic's $518 billion compute bets, which include a large Broadcom-linked equipment lease. Labs that want their own silicon need design capacity, and design talent is scarce.

Others are working on the same problem. In July, Synopsys itself showed a long-running chip verification agent built with Nvidia technology, per R&D World. So GPT-Synopsys is one entry in a crowded field, not the first.

What is still missing

The announcement leaves out the things buyers will ask about first. R&D World lists them: no general availability date, no numerical results for GPT-Synopsys, no pricing and no revenue split. Xenospectrum adds that no chip designed with the model has been shown passing sign-off and going into production. PromptZone also notes that some Hacker News commenters compared it with earlier Synopsys AI features from 2024 and 2025, and that no public benchmarks accompanied the launch.

Until those exist, GPT-Synopsys is a plan. The real test will be a customer case showing fewer iterations or a shorter schedule on an actual design. Watch for that, and for whether the revenue-sharing structure holds up once real customers are involved.

FAQ

What is GPT-Synopsys?

It is a specialized AI model being built by OpenAI and Synopsys to operate Synopsys' chip design software, interpret results and iteratively improve designs. It was announced on September 30, 2026.

Will AI now design chips on its own?

No. Synopsys says the model's output will still be checked by conventional verification tools, and the companies have not claimed it can take a design to manufacturing without human involvement.

How will OpenAI and Synopsys make money from it?

OpenAI pays Synopsys a training subscription fee and licenses its tools. After customers adopt it, the companies split revenue based on how much the model improves chip designs. The split itself has not been disclosed.

When can customers use it?

No date has been announced. The companies have not published pricing or benchmark results either.