Nvidia used AI to slash a chip‑design task
Nvidia says an internal language model trimmed an eight‑engineer, ten‑month GPU design task to an overnight run on a single GPU, though the company cautioned AI remains far from replacing human designers (tomshardware.com). The claim underscores how AI is being used as a force multiplier inside advanced engineering teams rather than as a full automation of chip development (tomshardware.com).
Designing a modern graphics processing unit is like planning a city of tiny switches, and Nvidia says one internal artificial intelligence model now handles one chip-design job overnight. (tomshardware.com) Tom’s Hardware reported on April 14, 2026 that Nvidia said the task had previously taken eight engineers about 10 months and now runs on a single graphics processing unit in one night. The outlet said Nvidia also warned the industry is still “a long way” from artificial intelligence designing chips without human input. (tomshardware.com) The work sits inside electronic design automation, the software layer chip companies use to draw circuits, test timing, place components, and catch errors before manufacturing. Nvidia’s design automation research group says it works across the flow from register-transfer level design to verification, physical design, sign-off, and design-for-manufacturing. (research.nvidia.com) Nvidia has been building these tools for years, not weeks. In October 2023, the company said its custom internal large language model, ChipNeMo, was trained on internal data to generate and optimize software and to assist human designers with a chatbot, a code generator, and a bug-analysis tool. (blogs.nvidia.com) By February 2025, Nvidia researchers were describing a broader “Marco” framework of multiple artificial intelligence agents for chip-design tasks such as generating Verilog hardware code, fixing syntax errors, and debugging timing problems. Nvidia said one timing-analysis agent delivered about a 60-times speedup versus human engineers. (developer.nvidia.com) That pattern helps explain the new claim: Nvidia is using language models less as a push-button chip designer and more as a specialist assistant for narrow, repetitive jobs inside a much larger engineering process. Nvidia’s own research group says it trains custom models for chip design and develops systems that “leverage” them for specific tasks. (research.nvidia.com; blogs.nvidia.com) The company has also been pairing artificial intelligence with graphics processing units in older parts of the toolchain. In March 2023, Nvidia said its AutoDMP system used graphics processing units and machine learning to improve macro placement, a floor-planning step that affects power, performance, and area. (developer.nvidia.com) Nvidia researcher Haoxing “Mark” Ren, who leads the design automation group, has framed the effort as a productivity push inside chip teams rather than a replacement plan. His author page says he leads Nvidia’s Design Automation research group and focuses on artificial intelligence for electronic design automation and graphics processing unit-accelerated tools. (developer.nvidia.com) So the immediate change is not that artificial intelligence can ship a graphics processing unit by itself. It is that Nvidia says one of the slow, specialized jobs inside chip design has moved from an eight-person, 10-month effort to a one-night run, with human engineers still in charge of the rest. (tomshardware.com); (research.nvidia.com)
