Nvidia’s Feynman GPU plan may tap Intel Foundry for I/O and packaging, hinting at a broader supply shift

Nvidia is expected to source select components for its next-generation Feynman GPUs from Intel Foundry, according to a DigiTimes report, marking a notable potential win for Intel’s contract manufacturing business.

The move would not replace TSMC as Nvidia’s primary chipmaker, but it could diversify key parts of the supply chain.

The report says Intel could manufacture I/O dies for Feynman using either its 18A or 14A process technologies, with the main GPU compute die still slated for TSMC on an advanced node. That split approach reflects how modern high-performance chips increasingly combine multiple dies optimized for different functions.

Packaging becomes part of the strategy

Beyond wafer production, Nvidia is also said to be considering Intel’s EMIB advanced packaging for some Feynman configurations. At the same time, much of the lineup would continue to rely on TSMC’s CoWoS packaging, which has been central to today’s high-bandwidth memory and AI accelerator designs.

DigiTimes estimates a roughly 75–25 division between TSMC and Intel across relevant packaging and associated work, though the exact mix could vary by product tier and customer demand. Both EMIB and CoWoS aim to connect multiple chiplets with high bandwidth and lower power, a key advantage as GPUs scale.

Why Intel Foundry is in play?

Intel has been positioning 14A as a major milestone for Intel Foundry, following 18A, with industry attention focused on whether it can attract top-tier external customers at scale.

For Nvidia, shifting I/O dies or packaging to Intel could help manage capacity constraints and reduce overreliance on a single supplier.

The report does not corroborate earlier rumors that suggested some lower-end Nvidia products might be built on Intel 14A, keeping the focus instead on Feynman-era components. If the collaboration materializes, it would underscore how competition in GPUs is increasingly tied to foundry access, packaging capacity, and multi-supplier execution.