Nvidia's Feynman Chip Ushers in the Angstrom Era of AI Computing

Nvidia Sets the Stage for Feynman at GTC 2026

Nvidia is preparing to pull the curtain back on Feynman, its next-generation AI chip architecture, at the GTC 2026 developer conference running March 16-19 in San Jose, California. CEO Jensen Huang teased the reveal during an impromptu dinner with engineers from Nvidia and SK hynix at a Korean fried chicken restaurant in Santa Clara, telling reporters: "We've prepared several new chips the world has never seen before."

A Leap Into Angstrom-Class Manufacturing

Feynman represents a landmark moment in semiconductor manufacturing. The architecture will be built on TSMC's A16 node, a 1.6-nanometre-class process that enters what the industry calls the "angstrom era." For the first time in Nvidia's product line, A16 introduces back-side power delivery through a technology called Super Power Rail, which routes electrical power through the back of the chip rather than competing for space with data signals on the front. The result is an eight to ten percent speed improvement or a fifteen to twenty percent reduction in power consumption compared to TSMC's previous N2P node.

Dual-Foundry Strategy Breaks New Ground

In a historic shift, Nvidia is pursuing a dual-foundry approach for Feynman. While TSMC will manufacture the high-performance compute dies, Intel has been tapped to produce the I/O dies using its 14A process node and EMIB advanced packaging technology. Nvidia invested five billion dollars in Intel in September 2025 to underpin this diversification strategy. Intel is expected to handle roughly twenty-five percent of total Feynman production, with risk production and pilot runs beginning throughout 2026 and 2027.

Memory Partners Rally Around GTC

Samsung and SK hynix will both present at GTC, showcasing their sixth-generation HBM4 memory designed for Nvidia's Rubin GPUs. Samsung began mass-production shipments of HBM4 on February 12, while SK hynix is expected to supply approximately two-thirds of Nvidia's total HBM4 volume this year. Samsung also plans to preview its next-generation HBM4E technology and its compatibility with future Nvidia architectures, potentially including Feynman.

The Road to 2028

Feynman is targeting a full launch in 2028, succeeding the current Blackwell and upcoming Rubin architectures. TSMC is set to begin mass production on the A16 process in late 2026, giving Nvidia a clear runway toward production readiness. The chip was first disclosed on Nvidia's roadmap at GTC 2025, continuing the company's pattern of announcing future architectures years ahead of launch.