 |
| Intel Foundry could help Nvidia with its Feynman GPUs |
For years, Intel’s foundry ambitions felt like a classic case of "wait till next year." The much-hyped 18A node came and went, largely shrugged off by external consumers who saw little reason to shift from TSMC’s iron grip on the market. But the narrative is quietly changing. Behind the scenes, whispers about Intel’s next big bet—the 14A node—are growing louder, with industry insiders calling it ‘the real deal.’
And now, we have a clearer picture of who might be lining up. The list of prospective clients reads like a who’s who of Silicon Valley: Apple, surprisingly AMD, and even the current king of AI, Nvidia. However, a deep dive into a new DigiTimes report reveals that not all these partnerships are created equal. For some, the collaboration is less about full commitment and more about strategic signaling.
Nvidia’s Feynman GPUs: A Toehold for Intel, Not a Takeover
Let’s start with the most headline-grabbing name: Nvidia. According to supply chain sources speaking to DigiTimes, Nvidia’s 2028-bound Feynman GPUs will indeed use Intel silicon—but only for the supporting cast. The plan is to manufacture the I/O dies on either Intel 18A or the more promising 14A node, depending on final yields. The primary GPU compute die, the heart of the AI powerhouse, will continue to rely on TSMC’s cutting-edge nodes, likely a custom N3 or N2 variant.
This is a classic "have your cake and eat it too" strategy. Nvidia gets to appease political pressure for "US-made" components and test-drive Intel’s new process without risking its core performance crown.
The packaging story is similar. Nvidia plans to incorporate Intel’s EMIB (Embedded Multi-die Interconnect Bridge) technology into some chips—a nice win for Intel’s advanced packaging ambitions. But the overwhelming majority of Feynman’s silicon will still use TSMC’s mature CoWoS (Chip-on-Wafer-on-Substrate) packaging. The estimated split is a decisive 75-25 in favor of TSMC.
What does this mean for the rumor that entry-level Nvidia products (like a successor to the N1X) could be fully fabricated on Intel 14A? The DigiTimes report pours cold water on that idea. It makes no such observation, instead focusing on the limited, specialized role for Intel.
The Apple Puzzle and the AMD Wildcard
If Nvidia’s engagement is cautious, Apple’s appears more… strategic. The report reiterates earlier leaks suggesting that future Apple M-series silicon could be based on Intel 14A—but with a significant asterisk. This would likely be for the base-level, non-Pro M-series chips found in entry-level MacBooks.
The high-end Pro, Max, and Ultra variants, along with the A-series chips for iPhones, will resolutely stick with TSMC. For Apple, this looks like a hedging move: maintain a relationship with an alternative US-based foundry for lower-risk components, while keeping the crown jewels safely inside TSMC’s fabs.
And then there’s AMD—the name that makes this whole story puzzling. AMD and Intel are arch-rivals in the x86 CPU space. The idea of AMD handing over any of its wafer orders to Intel Foundry is almost surreal. Yet, the report lists AMD as a prospective client. While no product details are provided, it suggests that the pressure for supply chain diversification and "US-first" manufacturing is so immense that even bitter rivals are being forced to the same negotiating table.
Why Now? The Political and Practical Shove
So, why are Apple, Nvidia, and even AMD suddenly cozying up to Intel Foundry after years of ignoring 18A?
The answer has three layers, all stemming from one source: geopolitics and risk management.
- The "America First" Mandate: With potential tariffs and subsidies tied to US manufacturing, tech giants are scrambling to show they have a domestic alternative to TSMC’s Arizona fabs.
- The Anti-Monopoly Move: TSMC’s dominance has become a systemic risk. Regulators in the US and Europe are uncomfortable with a single company controlling over 90% of the world’s most advanced chips. Handing Intel a few token orders helps ease that antitrust pressure.
- Cost and Capacity: TSMC’s advanced nodes are expensive and often oversubscribed. Intel 14A, if it truly is "the real deal," offers a potential second source for non-critical dies, giving Nvidia, Apple, and AMD more negotiating leverage.
The bottom line? Intel Foundry's 14A node is no longer a fantasy. It has real, world-class prospects. But for now, the biggest names in tech are treating it like a reliable backup singer, not the lead vocalist. The headliner for high-performance compute, at least through 2028, remains unmistakably TSMC.