OpenAI built an AI that can actually run the software chip engineers use

GPT-Synopsys combines OpenAI's frontier AI with Synopsys' electronic design automation (EDA) tools, the specialized software semiconductor companies use to design and verify chips. The model is built to reason about chip design and verification, and to directly operate Synopsys' own tools, rather than just offering suggestions for a human to carry out manually.

Why it matters: most AI coding or design assistants stop at generating suggestions, leaving a person to actually execute them inside the real software. GPT-Synopsys is built to close that gap entirely for chip design specifically, running the tools itself rather than describing what an engineer should click next.

Engineers delegate a design objective, things like power, performance, and area (PPA) optimization, timing closure, or verification, to the model. From there, GPT-Synopsys runs the relevant tools, interprets the results, implements changes, and iterates toward a verified outcome that an engineer then reviews and approves.

• Engineer's role: set the design objective, review and approve final outputs.
• Model's role: operate the tools, interpret results, make changes, iterate.
• Integration: built into Synopsys.ai and the Synopsys Autopilot platform.
• Infrastructure: runs on OpenAI-hosted servers.

In real life it's the difference between a junior engineer asking a senior colleague for advice on what to try next, and actually handing the senior colleague the keyboard to run the next dozen experiments themselves while you focus on reviewing what comes back.

The catch: letting an AI model directly operate the software behind a company's unreleased chip designs raises obvious sensitivity questions. Synopsys and OpenAI addressed that directly: customer design data won't be used to train the model, will be encrypted at rest and in transit, and comes with configurable retention, audit, and permission controls.

Why it matters: chip designs represent some of the most closely guarded intellectual property in the tech industry, often years of R&D investment. Those specific privacy commitments are a precondition for any semiconductor company to even consider running real, unreleased designs through a third-party AI system at all.

It's a multi-year strategic partnership. OpenAI is licensing Synopsys' EDA tools to build and operate the model, and the two companies will share revenue from GPT-Synopsys while jointly marketing it to customers worldwide. The offering bundles compute, the model itself, and the necessary software licenses together into one package.

Why it matters: that structure means neither company is simply reselling the other's product, they're building genuinely combined infrastructure and splitting the resulting revenue, a deeper commitment than a typical technology partnership or simple API integration.

Synopsys CEO Sassine Ghazi framed the partnership as "bringing frontier intelligence to chip design to help more companies develop" advanced silicon while maintaining the manufacturing rigor the industry requires. OpenAI co-founder Greg Brockman put it more simply: "With Synopsys, we're bringing that work to chip design, helping engineers explore more designs and get to working chips faster."

Background: no. OpenAI has separately partnered with Broadcom on its own AI-specific chips, including an inference chip called Jalapeno, which has reportedly outperformed comparable specialized processors from competitors. That earlier work focused on chips OpenAI itself would use to run its own AI models.

Why it matters: GPT-Synopsys is a different kind of hardware-adjacent bet, not a chip OpenAI uses internally, but an AI tool aimed at the broader semiconductor industry's own design process. Together, the two efforts show OpenAI investing in chip-related work from both directions: the hardware its models run on, and the tools other companies use to design hardware of their own.

Why it matters: modern chip design involves an enormous amount of complexity, exploring countless possible configurations to balance power, performance, and area, then verifying each option actually works correctly before manufacturing.

That combination of vast possibility space and mechanically verifiable correctness is exactly the kind of problem where an AI system that can run tools and check its own results tends to perform well, closer to how AI has already proven useful in domains like protein folding than in more open-ended creative tasks.

Background: Synopsys has been a dominant force in electronic design automation since introducing the first commercially successful logic synthesis tool back in 1986. Today, the EDA market is controlled almost entirely by just a handful of companies: Synopsys, Cadence Design Systems, and Siemens EDA, who together hold roughly 75% of the overall market, up from about 45% in the early 1990s as the industry consolidated around fewer, larger players.

Synopsys also recently moved to expand that position further, announcing a roughly $35 billion acquisition of engineering simulation company Ansys in January 2024, a deal structured as cash plus stock for each Ansys share. That acquisition signals Synopsys building out its capabilities well beyond pure chip design software, into the broader engineering simulation tools companies use alongside chip design itself.

Why it matters: a company with that much control over the actual tools chip engineers use industry-wide is in a uniquely powerful position to embed AI directly into real, existing workflows, rather than building a separate tool engineers would need to adopt from scratch. That's a meaningfully different starting point than an AI startup trying to convince an entire industry to switch to a brand-new platform from zero.

It also means GPT-Synopsys has a plausible path to reaching a large share of the semiconductor industry relatively quickly, simply because so many chip companies already run their design work through Synopsys tools day to day.

• Faster chip design cycles could eventually shorten the wait between announcing new hardware and actually being able to buy it, though that effect would take years to show up clearly.
• This won't show up in a consumer product directly, but it could speed up how quickly new chips, including ones powering future consumer devices, actually reach the market.
• If you work in semiconductor design, early access is already rolling out to select customers. Watch for broader availability announcements as engagements expand.
• Customer data protections are explicitly built in, encryption, no training use, audit controls, worth understanding if your company is evaluating the tool.
• This is a template worth watching beyond chips. An AI that directly operates existing professional software, rather than just chatting about it, is a pattern likely to spread into other specialized engineering fields.