NVIDIA Explores Intel’s 18A/14A Process and EMIB Packaging for Future Feynman AI Chips, Indicating a Significant Move Away from TSMC

NVIDIA is reportedly investigating strategies to broaden its chip supply chain, with Intel Foundry emerging as a promising partner for its upcoming AI product lines.

NVIDIA Explores a “Low Risk”Collaboration with Intel Foundry for Feynman I/O Die

As TSMC begins to showcase significant bottlenecks within the AI chip supply chain, fabless manufacturers are actively seeking ways to broaden their sourcing. Intel’s recent advancements with its 18A process have created momentum in its foundry division, potentially attracting external clients. According to a report from DigiTimes, NVIDIA is now considering utilizing Intel’s 14A and 18A processes specifically for its Feynman I/O die, coupled with the advanced EMIB packaging technology.

Reports indicate that NVIDIA does not intend to fully commit to Intel Foundry for the Feynman project, instead opting to source from both Intel and TSMC. Specifically, NVIDIA plans to outsource only the Feynman die to Intel, likely favoring the 14A node due to its feasibility. Furthermore, Intel is anticipated to provide its EMIB technology for the Feynman production, which will see Intel responsible for 25% of the overall production, while TSMC takes on the remainder.

Man in Intel Foundry vest holding a silicon wafer in front of an Intel sign outdoors. — Intel’s CEO showcasing an 18A wafer | Image Credits: Intel

A number of American fabless manufacturers, including AMD and Qualcomm, are adopting a dual-foundry strategy by partnering with both Samsung and TSMC. This trend is primarily driven by two critical issues: the reliance on TSMC for both front-end and back-end manufacturing creates a potential risk amid rising geopolitical tensions between China and Taiwan. Moreover, the immense scale of AI development is causing manufacturers and cloud service providers to rush for TSMC’s limited capacity, leaving many unable to secure sufficient supplies.

NVIDIA’s strategy with the Feynman project is characterized as “low risk”because assigning the I/O die to Intel Foundry safeguards against potential setbacks related to yield or capacity issues. This arrangement also allows NVIDIA the flexibility to outsource non-core products to Intel, potentially including next-gen gaming GPUs in future collaborations.

It will be interesting to observe how the dynamics at Intel Foundry progress in the coming months, especially as TSMC’s clients increasingly endeavor to diversify their chip supply sources, with Intel and Samsung positioned as the leading alternative options.

Source & Images