Following Apple's lead, NVIDIA also plans to shift part of its chip manufacturing business to Intel's foundry services. NVIDIA's Feynman architecture platform, expected to launch in 2028, will be developed in cooperation with Intel, marking the latest case of a US tech giant adjusting its supply chain strategy under the Trump administration's push for a "Made in the USA" goal.

According to a DIGITIMES report, supply chain sources revealed that NVIDIA will adopt a "low-volume, low-end, non-core" cooperation strategy with Intel. The core GPU chips will still be manufactured by TSMC, while the I/O chips will partially utilize Intel's 18A process or the 14A process slated for mass production in 2028, with final advanced packaging handled by Intel's EMIB technology. Based on the proportion of advanced packaging, Intel's share could reach up to 25%, with TSMC accounting for approximately 75%.

This shift reflects how US tech companies, facing political pressure, tariff threats, and supply chain resilience considerations, are compelled to move from a model highly concentrated on TSMC towards a new strategy of "multi-sourcing and risk diversification." Besides NVIDIA and Apple, companies including Google, Microsoft, AWS, Qualcomm, Broadcom, AMD, and Tesla are also in talks with Intel about potential cooperation.

Although some orders are being diverted to Intel, industry observers believe this is "far more beneficial than harmful" for TSMC, as it helps alleviate monopoly concerns, releases political pressure, while TSMC remains confident in securing major high-end chip manufacturing contracts.

NVIDIA's Feynman architecture will partially shift to Intel. After announcing a $5 billion investment in Intel in September 2025, NVIDIA's latest plan involves cooperating with Intel on the Feynman architecture chips, the successor to the Rubin series. Supply chain information indicates that core GPU chips remain with TSMC, while I/O chips will partially use Intel's 18A or 14A process, with the specific choice depending on the subsequent yield and mass production status of the 14A process.

Supply chain sources stated that under the established "Made in the USA" goal and tariff pressures from the Trump administration, major US chipmakers have long been discussing cooperation with Intel. However, as the 18A process did not meet client expectations, the cooperation timeline is likely aligned with the mass production of the 14A process in 2028. Intel CEO Pat Gelsinger recently mentioned that two clients are currently evaluating the specific details of the 14A process.

Given the high risks associated with adopting the 14A and 18A processes, most companies are starting cooperation with Intel through advanced EMIB packaging first. Supply chain analysts note that while the "Made in the USA" goal faces cost and yield challenges, the realities of politics, supply chain resilience, and constraints on TSMC's advanced packaging capacity will inevitably push US chipmakers to adopt a dual-foundry strategy.

Apple is restarting cooperation with Intel for its entry-level processors. The product under discussion for some time between Apple and Intel is likely the "entry-level M-series processors" used in MacBooks, currently manufactured by TSMC. Apple's Mac series began using Intel x86 processors in 2006, and Intel even established a dedicated "Apple Group" production line at its Oregon fab in the US specifically for Apple in 2005.

In June 2020, Apple officially announced its Arm-based "Apple Silicon" chips, transitioning the entire Mac lineup to its own chips within two years. The primary considerations at the time were to increase supply chain control and integrate its ecosystem, with another key factor being Intel's delays in its 10nm process potentially impacting new Mac launches.

Supply chain sources reveal that the main reason for Apple restarting cooperation with Intel three years later remains the "Made in the USA" goal and tariff impacts under Trump's leadership, followed by factors like cost, diversifying single-source manufacturing risks, and capacity shortages.

TSMC's three-tier strategy to handle client diversification. Currently, both Apple and NVIDIA are gradually adjusting with products that carry the lowest manufacturing risk. Other companies in talks with Intel include Google, Microsoft, AWS, Qualcomm, Broadcom, AMD, and Tesla, along with highly secured long-term contracts for US government projects. However, significant uncertainty remains regarding whether Intel can meet the demands of tech giants already accustomed to TSMC's model.

For TSMC, it has foreseen that many clients will place some orders with Intel, but analysis suggests this is actually "far more beneficial than harmful," with at least three strategic considerations: First, it can reduce monopoly and regulatory concerns; Second, it can alleviate US political pressure; Third, the spillover involves only "non-core" orders, which could strengthen TSMC's future bargaining power and supply position.

On one hand, this helps resolve potential antitrust concerns related to TSMC's overwhelmingly high market share in the foundry industry. On another, appropriately releasing non-core orders can alleviate pressure from the Trump administration's various demands. Furthermore, TSMC remains confident in securing major contracts for high-end chips, which are the core business for these large companies. Perhaps clients who try diverting orders to other foundries will come to appreciate TSMC's advantages, which could ultimately strengthen TSMC's position in future pricing and supply negotiations.