The news, reported by Reuters and attributed to six people familiar with the matter, marks a turning point: Synopsys, the American giant of electronic design automation (EDA), is preparing to exit the business of manufacturing control software for semiconductor fabs. The engineers working in that area will be redirected toward a far more lucrative segment: the design of chips optimized for artificial intelligence.

The decision is more than an internal reshuffling. It reveals a structural rift in how the semiconductor industry is reallocating its value. Fab software—which handles yield management, defect control, and lithography processes—is critical for running foundries like TSMC or Samsung. But it has become a low-growth commodity, squeezed by process maturity and competition from other vendors such as Applied Materials and KLA.

By contrast, demand for AI chips—both for data center training and edge inference—is exploding. Designing a custom accelerator requires simulation and synthesis tools that must be rethought for unconventional architectures (near-memory compute, optical networking, hardware sparsity). This is a frontier promising high margins and multi-year contracts, especially from major cloud providers and startups challenging Nvidia.

Synopsys’ shift makes clear who wins and who loses. AI chip designers will gain from more powerful, focused tools, accelerating time-to-market. Fabs, on the other hand, risk seeing reduced innovation in control software, with a heavier impact on less advanced production lines—those that cannot afford in-house development teams. Over the long term, the gap between leading-edge process nodes and mature ones could widen, with geopolitical implications for manufacturing sovereignty.

But there is a third consequence, less immediate yet crucial for those tracking hardware for on-premise inference. If AI chip design becomes faster and cheaper, the market may soon see a wave of new efficient accelerators for running LLMs locally. Today, on-premise deployment of large models faces cost and power barriers tied to GPUs. Tomorrow, chips designed with verticalized tools could lower the VRAM and TCO hurdle, making data sovereignty more tangible without sacrificing performance. It’s not a guaranteed future, but the direction is clear.

In this scenario, Synopsys positions itself as the enabler of a more fragmented, competitive ecosystem. It’s a bold bet: exiting a consolidated market to enter one still fluid, where today’s alliances (with Nvidia, AMD, Intel) could turn into direct competition tomorrow, as customers themselves become silicon designers. The game is on, and the signal to the industry is unmistakable: value is moving upstream, where software designs the intelligence of silicon.