NVIDIA could use Intel’s Foveros advanced packaging technology for some of its future artificial intelligence products during the second half of 2028, according to a forecast attributed to analyst Jeff Pu of GF Securities. The possibility would be a new opportunity for Intel’s manufacturing business, which is looking to win clients among major semiconductor companies, although there is currently no public confirmation of a deal between the two companies.
Intel Foveros and NVIDIA in 30 seconds
- Analyst Jeff Pu suggests NVIDIA could turn to Intel for the packaging of some AI chips in the second half of 2028.
- The technology in question is Foveros, which vertically stacks different chip components to reduce the footprint of the package.
- The forecast coincides with the timeline set for Intel 14A production, although using Foveros doesn’t confirm that NVIDIA will manufacture its GPUs with that process.
- Intel is looking to expand its foundry and advanced packaging business against TSMC, which holds a leading position in manufacturing advanced chips.
- The possible collaboration remains an analyst forecast, with no public details on products, volumes, or commercial terms.
The opportunity for Intel lies in a stage of manufacturing that has gained importance as AI accelerators grow increasingly complex. Beyond manufacturing the main circuits, companies need to integrate different components into a single package, with connections capable of moving data quickly while keeping power consumption under control.
NVIDIA, one of the leading providers of processors for artificial intelligence, needs advanced integration technologies for its future generations of chips. According to information published by Wccftech, the company could be looking at Intel solutions for some of these tasks while keeping TSMC as the likely manufacturer of the main components. The actual scope of that potential collaboration is still undefined.
Foveros opens a path for Intel in advanced packaging
Intel has long been trying to expand its activity as a manufacturer for third parties. While it competes to gain a foothold in the most advanced production processes, it is also developing packaging technologies that allow several blocks of silicon to be integrated into a single product.
These include EMIB and EMIB-T, which use interconnect bridges to link components placed side by side. These solutions have come up in earlier reports about Google’s possible interest in Intel technologies for its tensor processing units (TPUs), used in artificial intelligence workloads. However, the information cited by Wccftech shouldn’t be confused with contractual confirmation of any of those projects either.
Foveros takes a different approach. Instead of simply placing components side by side, it allows different blocks of silicon, including compute components, to be stacked vertically on a base. This three-dimensional arrangement can reduce the horizontal footprint of the package and shorten certain connections between the integrated elements.
The physical proximity of the components can help with bandwidth and power efficiency, although the actual results depend on the design, the manufacturing technology, and the operating conditions. Stacking also raises thermal and manufacturing challenges, so its appeal has to be assessed product by product.
For Intel, getting NVIDIA to use Foveros would carry significant commercial value. It wouldn’t necessarily mean NVIDIA shifting production of an entire GPU to Intel, but it could open a path for collaboration on a relevant part of the manufacturing chain. In AI chips, where accelerators can combine several compute and memory blocks, packaging is a decisive stage for achieving the expected performance.
The possible collaboration coincides with Intel’s 14A timeline
Jeff Pu’s forecast places a possible use of Foveros by NVIDIA in the second half of 2028. That timeframe coincides with the forecasts published for the Intel 14A manufacturing process, whose high-volume production has also been placed in that period.
However, the two things are different. Intel 14A is a semiconductor manufacturing node, while Foveros is a three-dimensional packaging and integration technology. NVIDIA eventually using Foveros wouldn’t by itself prove that the company will produce its main GPUs on Intel 14A.
Back in January, reports had already circulated suggesting NVIDIA could use Intel processes for certain components of future Feynman GPUs, a generation that has been linked to the 2028 timeframe. Those reports raised the possibility of keeping production of the main compute block at TSMC while turning to Intel for auxiliary components or certain design elements.
The new report attributed to Pu seems to fit that hypothesis, but it doesn’t provide enough detail to determine which components Intel might manufacture or package. Nor does it specify which NVIDIA product would use Foveros, how many chips would be involved, or whether there would be a commercial production commitment.
The distinction matters because manufacturing a GPU’s main compute block and handling its integration into a package are different activities. A manufacturer can take part in one stage without taking on the others. For NVIDIA, combining suppliers would let it explore industrial alternatives, but it would also require coordinating designs, validation processes, and production schedules.
Intel looks for clients while TSMC absorbs AI demand
The rise of artificial intelligence has increased pressure on the industrial capacity needed to manufacture and assemble accelerators. Production depends on several stages, from wafer fabrication to chip dicing, integration, and final testing. Bottlenecks at any of them can affect the availability of the finished product.
Against that backdrop, Intel is trying to turn its packaging technologies into an additional source of business. The company doesn’t need to fully displace TSMC to compete in the market: it could compete for certain manufacturing or integration stages, as long as its solutions meet each customer’s requirements for performance, cost, and availability.
For NVIDIA, any potential collaboration would need to be assessed on technical and commercial grounds. Integrating components from different suppliers can expand the available options, but it also requires validating design compatibility, thermal performance, reliability, and delivery timelines.
For now, the scenario described by Wccftech remains a forecast based on analyst information. No details have been published that would confirm NVIDIA has officially selected Foveros for a specific generation of products. Nor is there any confirmation in the cited information of volumes, prices, or production commitments.
The second half of 2028 is, therefore, a timeframe for a possible collaboration, not a confirmed launch date. Until the companies announce specific deals or products, the potential arrangement should be understood as an opportunity for Intel within a market where advanced packaging is gaining weight alongside chip manufacturing itself.
Frequently asked questions
What is Intel Foveros?
Foveros is an advanced packaging technology that allows different silicon components to be stacked vertically. Its design can reduce the footprint of the package and improve communication between the integrated blocks.
Will NVIDIA manufacture its GPUs at Intel?
That’s not confirmed. The information cited points to a possible use of Foveros for packaging tasks, but it doesn’t show that NVIDIA will shift manufacturing of its main GPUs to Intel.
When could NVIDIA use Foveros technology?
The forecast attributed to analyst Jeff Pu places that possibility in the second half of 2028. There is no official date or confirmed product from NVIDIA.
What’s the difference between Foveros and Intel 14A?
Foveros is a packaging technology that integrates several components into a single assembly, while Intel 14A is a semiconductor manufacturing process. A packaging collaboration doesn’t necessarily mean the main chip will be manufactured on Intel 14A.
via: wccftech
