Nvidia considers Intel Foundry for Feynman AI chips as TSMC capacity constraints bite

Nvidia Explores Intel Foundry Partnership for Future AI Chips

Nvidia is reportedly in talks with Intel Foundry to manufacture components of its upcoming Feynman AI chips, according to DigiTimes citing supply chain sources1

1

2

2

. This potential collaboration marks a significant departure from Nvidia's traditional reliance on TSMC and represents a calculated attempt to diversify supply chain amid mounting capacity constraints in the semiconductor manufacturing industry. The deal, expected to materialize around 2028, would see Intel handling non-core portions of the next-generation GPU while TSMC continues producing the critical compute dies3

3

.

Strategic Allocation Between Intel 18A Process and TSMC

The arrangement follows a "low risk" approach where Nvidia would use either the Intel 18A process or Intel 14A process specifically for the I/O die components of Feynman AI chips2

2

. The I/O die handles memory controllers and inter-chip connectivity, making it less performance-critical than the GPU compute die but still requiring advanced process nodes3

3

. TSMC will retain manufacturing of the most complex silicon using its A16 process node. Intel's selection depends heavily on whether the foundry business achieves sufficient yields on its cutting-edge technologies, with Intel 14A still facing uncertain production timelines1

1

.

EMIB Packaging Technology Secures 25% Share

Beyond wafer fabrication, Intel will provide advanced packaging services using its proprietary EMIB packaging technology for approximately 25% of Feynman production1

1

2

2

. EMIB connects various dies together on chips, a critical capability as AI accelerators grow increasingly complex. During Intel's fourth-quarter earnings call, CFO David Zinsner revealed that individual advanced packaging deals could exceed $1 billion, far surpassing earlier expectations of just a few hundred million dollars3

3

. This packaging opportunity may serve as Intel's entry point with major customers before expanding into broader semiconductor manufacturing relationships.

Supply Chain Risk Drives Dual-Foundry Strategy

The potential partnership reflects broader industry concerns about supply chain risk concentrated at TSMC. The Taiwan-based foundry currently dominates both front-end and back-end production, creating what industry observers describe as a "single point of failure" amid geopolitical tensions2

2

. TSMC's capacity constraints have left manufacturers scrambling to secure production slots during the AI buildout, with demand for CoWoS advanced packaging services outpacing supply3

3

. AMD and Qualcomm are similarly pursuing dual-foundry strategy approaches, pairing Samsung with TSMC to reduce dependency2

2

.

Political Pressure Accelerates US Manufacturing Push

Political pressure from the White House to boost US manufacturing has made once-unthinkable partnerships more viable1

1

. Both the US government and Nvidia have acquired stakes in Intel's foundry business, aligning financial and strategic interests. Intel CFO David Zinsner confirmed the company is "aggressively getting tools" ready for Intel 7, 10, 3, and 18A nodes but "holding back on 14A" until securing committed foundry customers1

1

. CEO Lip Bu-Tan warned in an SEC filing that without significant external customers for Intel 14A, the company might pause development of that node and successor leading-edge technologies.

Feynman Architecture Timeline and Implications

Feynman represents Nvidia's architecture following Rubin and Rubin Ultra, scheduled to arrive from 2028 onward. Jensen Huang revealed the roadmap at GTC, naming the architecture after physicist Richard Feynman, though technical specifications remain scarce1

1

. The chip pairs with Vera CPU, IO, and network components as part of Nvidia's vision for "Gigawatt AI Factories." With gaming GPUs still on the Blackwell architecture preceding Rubin, consumer products won't see these manufacturing changes immediately. However, reports suggest Nvidia may also outsource non-core gaming products to Intel Foundry in future arrangements2

2

.

Intel Foundry's Uncertain Path Forward

Recent hiring activity at Intel's Ohio fab complex, designated for Intel 14A production, suggests growing confidence in securing orders3

3

. Analysts also expect Apple to potentially use Intel 18A for M-series chips and Intel 14A for future iPhone components. Yet Intel Foundry faces an uphill battle competing with TSMC's established yields and proven track record. The Nvidia deal, while tentatively positive, remains unconfirmed and could collapse if Intel fails to demonstrate competitive performance on its advanced process nodes. For now, 14A and subsequent technologies remain on shaky ground, with their viability tied directly to landing major external customers willing to take calculated risks on Intel's semiconductor manufacturing capabilities.