NVIDIA could switch to Intel's Foveros 3D packaging technology for future products in 2028, while continuing to source the GPU dies themselves from TSMC, according to a report by Softonic.

The claim, if confirmed, would mark a notable shift in NVIDIA's supply chain strategy. The company has relied on TSMC not only for silicon fabrication but also for the advanced packaging that bonds compute dies to high-bandwidth memory and substrates. Under the reported arrangement, that packaging role would move to Intel Foundry, with TSMC retaining wafer fabrication of the graphics processors.

What does Foveros packaging actually do?

Foveros is Intel's die-stacking technology. It lets manufacturers place logic dies on top of one another with vertical interconnects, shortening signal paths and increasing bandwidth density between chiplets. For GPU-class products, where memory bandwidth and die-to-die links set performance ceilings, packaging quality directly affects throughput and power efficiency.

Advanced packaging has become a bottleneck across the industry. Demand for CoWoS-class capacity at TSMC has outstripped supply for consecutive years, forcing chip designers to qualify additional suppliers. NVIDIA adding Intel as a packaging partner would give the company a second source for one of the scarcest steps in AI accelerator production.

Who makes what under the reported plan?

The split described in the report is straightforward:

• TSMC — fabrication of the GPU compute dies
• Intel Foundry — Foveros-based 3D packaging of those dies into finished modules
• 2028 — the timeframe in which the changeover could take effect

Such a division of labor is not unprecedented. Intel Foundry already packages silicon for external customers, and NVIDIA routinely splits production of a single product line across multiple fabs and assembly partners to secure capacity.

Why does 2028 matter?

A 2028 target aligns with the next generation of process nodes and product refresh cycles. By then, both TSMC and Intel expect to be running their most advanced lithography in volume, and packaging technologies on both sides — Foveros derivatives at Intel, CoWoS successors at TSMC — will have advanced accordingly.

For Intel, winning NVIDIA packaging orders would represent a commercial endorsement of its foundry ambitions from the largest AI chip maker by revenue. For NVIDIA, it would reduce single-supplier risk at the packaging stage, which analysts have repeatedly identified as a constraint on shipment growth.

The report does not state whether a contract has been signed, nor which specific NVIDIA product lines would use Foveros. Neither NVIDIA nor Intel has commented publicly on the claim, and 2028 remains a target date subject to qualification schedules and capacity availability.

For now, the story signals direction rather than a done deal: NVIDIA is evaluating a supply chain in which Intel handles the stacking and TSMC keeps the silicon.