Nvidia May Evaluate Intel Foveros Packaging in 2028

Highlights

A report by GF Securities analyst Jeff Pu says Nvidia may consider using Intel’s Foveros advanced packaging for some backend work in the second half of 2028. No contract or specific product decision has been confirmed, so the possibility remains market speculation. The discussion echoes a DigiTimes report from January suggesting that TSMC could continue manufacturing the main Feynman GPU dies, while Intel may be assessed for I/O or other non-core dies and packaging. Intel’s 14A process is expected to reach high-volume production in the second half of 2028. For Nvidia, Intel packaging could offer another way to ease reliance on constrained TSMC packaging capacity.

Sentiment Analysis

• The article’s overall tone is cautious and analytical. It describes a potential business opportunity but repeatedly stresses that Nvidia has not confirmed a contract or selected a specific product.


• The outlook is somewhat positive for Intel: a limited packaging role could validate Foveros at industrial scale and establish a foothold in the data-center AI supply chain. However, the report does not present the possibility as a guaranteed win.


• For Nvidia, the proposed arrangement is framed as a practical way to diversify packaging capacity and reduce bottleneck risks during periods of high AI accelerator demand. TSMC would nevertheless remain central to manufacturing the main GPU dies under the scenario described.


• The sentiment is therefore mixed but measured: potential benefits are balanced against uncertainty, limited scope, and the absence of confirmed agreements. The article focuses on supply-chain strategy rather than making claims about an immediate shift in technology leadership.

Article Text

A new report from GF Securities analyst Jeff Pu suggests that Nvidia may evaluate Intel’s Foveros advanced packaging technology for some backend operations in the second half of 2028. The report does not describe a signed agreement, a confirmed order, or a specific Nvidia product decision. The possibility should therefore be understood as market speculation, rather than an announced change to Nvidia’s manufacturing plans.

The discussion concerns advanced packaging, a stage of semiconductor production in which separate dies are connected and assembled into a finished package. Nvidia’s AI accelerators depend on complex manufacturing and packaging arrangements, and demand for these products has put pressure on available capacity. The reported possibility of working with Intel would add another potential supplier for some packaging tasks, but it does not imply that Nvidia is moving all GPU production away from its established partners.

The speculation is consistent with a report published by DigiTimes in January. That report said the main dies for Nvidia’s future Feynman-series GPUs could continue to be manufactured using TSMC’s advanced process technology. Intel might instead be considered for I/O dies or other non-core components, potentially using its 18A or 14A processes, with packaging also handled by Intel. These possibilities have not been confirmed by the companies.

Intel’s packaging portfolio includes EMIB, Foveros, and Foveros Direct. EMIB uses an embedded silicon bridge to connect chiplets horizontally. The EMIB-T approach adds through-silicon vias, which can support vertical power delivery and signal connections. Foveros takes a different approach by stacking compute dies vertically on a base die. This arrangement can reduce package area and shorten interconnect distances, with potential benefits for bandwidth and energy efficiency.

The timing is notable because Intel’s 14A process is expected to reach high-volume production in the second half of 2028, aligning with the period in which Nvidia may reportedly assess Foveros. The overlap in timing provides context for why Intel could be considered in a future product cycle. It does not, by itself, establish that Intel will receive an order or that its packaging will be selected for a Feynman product.

For Nvidia, using Intel for some packaging work could diversify reliance on TSMC’s CoWoS capacity and help ease potential bottlenecks when demand for AI accelerators is especially high. A multi-supplier strategy could give Nvidia additional options for assembling products, although the practical value would depend on technical qualification, manufacturing capacity, product requirements, and commercial terms. The report does not specify how much work Intel might receive.

For Intel, even a limited role in Feynman backend operations or a small number of non-core component orders could have strategic importance. The potential value would not necessarily lie in overtaking other companies in leading-edge front-end manufacturing in the near term. Instead, completing large-scale industrial validation of Foveros and entering the data-center AI supply chain could strengthen Intel’s position in advanced packaging.

The central point is that the reported scenario concerns a possible division of labor: TSMC could retain production of Nvidia’s main GPU dies, while Intel may be assessed for selected dies or backend packaging. That distinction matters when evaluating what the speculation could mean for competition. It points to potential supply-chain diversification, not a confirmed transfer of Nvidia’s core GPU manufacturing. Until the companies announce concrete decisions, the scope and commercial significance of any Intel role remain unknown.

Key Insights Table