GPT-Synopsys: OpenAI Gets Paid Only When the Chips It Helps Design Beat the Customer's Baseline

OpenAI and Synopsys announced a multiyear agreement on Sept. 30 to build GPT-Synopsys, a model meant to operate chip-design software directly. Synopsys said early engagements with leading semiconductor customers are under way, and named none of them. Synopsys' press release gives no release date, no price and no performance figure.

The terms that carry the news came from the company's Investor Day the same day. According to Futurum Group's summary, GPT-Synopsys is billed through three channels: tool subscriptions, consumption-based fees, and an outcome-based revenue share with OpenAI.

How the revenue share works

The revenue share is paid when the model achieves power, performance and area results beyond what a customer would reach without it, according to The Decoder and Futurum. Neither source gives the percentage.

Synopsys said engineers delegate design objectives such as power, performance and area optimisation, timing and verification. Agents then operate the tools, read the results and iterate toward verified outcomes. Engineers review and approve the output.

The model runs on OpenAI-hosted infrastructure. Customer data will not be used for training and is stored encrypted, The Decoder reported.

Item	What was disclosed	Source
Announced	Sept. 30, 2026	Synopsys
Term	Multiyear	Synopsys
Revenue split	Not given	Futurum, The Decoder
Release date	Not given	Synopsys
Named customers	None	Synopsys

What OpenAI is putting in

Futurum reports that OpenAI is investing hundreds of millions of dollars in post-training the model on Synopsys tools and workflows, and will handle security and data containerisation. That figure is Futurum's; the press release does not state it.

Sassine Ghazi, President and CEO of Synopsys, said the partnership will "expand access to Synopsys' advanced design capabilities". Greg Brockman, President and Co-founder of OpenAI, said the companies are "bringing that work to chip design, helping engineers explore more designs and get to a working chip faster."

The caveat Futurum puts in writing

Futurum lists a caution that the press release does not. Early GPT-Synopsys engagements are OpenAI-driven, and the analysis says outcome-based pricing "remains unproven" beyond that anchor customer. In other words, the revenue-share model has no external customer result behind it yet.

The same Investor Day set fiscal 2027 revenue guidance at a midpoint of $11.15 billion, 15% growth. Futurum notes that no new revenue stream appears in fiscal 2027 and that growth rests on conventional design-software demand.

Futurum also reports a $1 billion-plus, multi-year Synopsys contract with Amazon covering the Graviton, Trainium and Nitro processors, with per-unit royalties. Amazon is also the subject of our report on its plan to lease back $8 billion of Nvidia chips.

What to watch

OpenAI also builds its own inference chips with Broadcom, The Decoder reported, which makes it both a customer and a vendor of this class of tool. OpenAI has separately described a coordinated distillation campaign against its own models, so how it handles other firms' data and outputs is already a live question. Neither company has said when GPT-Synopsys reaches general availability or what the first named customer will report.