Nvidia is reshaping the very way semiconductors are developed by combining an in-house large language model (LLM) trained on decades of accumulated graphics processing unit (GPU) design data with Reinforcement Learning-based artificial intelligence (AI).
At Nvidia's GTC 2026, held on the 16th–17th, the company announced results showing that AI completed a "standard cell library porting" task—previously handled by eight engineers over about 10 months—in just one night. Beyond automation, the move is expected to have significant industry ripple effects because it compresses "development time," the largest expense in semiconductor design.
In semiconductor design, the "porting" task of rearranging existing layouts when shifting to a new process has long been cited as a representative bottleneck. As nodes shrink, design rules increase exponentially, making reliance on skilled engineers absolute.
To address this, Nvidia introduced NV-Cell, a design AI system based on Reinforcement Learning (RL). The AI learns tens of thousands of design rules and autonomously explores optimal cell placement routes. As a result, it not only dramatically reduced working time, but in some cases delivered improved outcomes over human designs in terms of area and performance.
Bill Dally, Nvidia's chief scientist, said in a conversation with Jeff Dean, Google's chief scientist, at GTC 2026 that "AI is being applied across the entire design process." However, noting clear technical limits, Dally said "there is still a long way to go before AI can independently design an entire chip." Analysts say that, for now, AI is focused less on fully autonomous design without human intervention and more on serving as an "intelligent assistant" that explosively boosts engineers' productivity.
In practice, Nvidia is training AI on architecture documents, error logs, and verification data generated in the GPU design process, accumulating design know-how as a data asset. As a result, the yardstick for semiconductor competitiveness is rapidly shifting from the personal experience of skilled engineers to the ability to leverage data and algorithms. Dedicated LLMs like "Chip Nemo" are serving as mentors for junior designers, leveling up design capabilities across the organization.
These changes are directly affecting production structures. Whereas it was once common to finish design completely before applying it to manufacturing, there is now a strengthening trend toward "design–process co-optimization," which reflects process conditions in real time from the design stage. Major foundry companies such as TSMC and Samsung Electronics are deeply involved from the design phase and refining collaborative structures in the same vein.
The adoption of AI is removing design bottlenecks and maximizing product launch speed. Competition in the semiconductor industry is expected to hinge on a "speed war" over how quickly companies can iterate and optimize the entire design and production cycle, beyond simple performance.