Intel’s foundry has spent years chasing a customer whose name would settle the question of whether the business can compete. A report this week holds out that customer, in the conditional tense that marks every leak of this kind.
Nvidia may adopt Intel’s Foveros chip-packaging technology in 2028, according to a report on October 9, the year the process is expected to reach mass production. Neither company has confirmed the report, and 2028 is far enough away that the word “may” is doing a great deal of work. But the possibility alone is a measure of how much the ground has shifted.
Foveros is Intel’s answer to the most important question in chips today: what happens when shrinking transistors gets harder. The technology stacks chips — or chiplets — on top of one another, bonding them face to face so that signals travel shorter distances and performance rises without a smaller process node. Intel introduced the first version in 2019 and has refined it since, selling it through its foundry business to chip designers who want to build with someone else’s manufacturing.
Packaging has quietly become the front line of the industry. Nvidia’s best-known designs already lean heavily on advanced packaging to bind together the processors and memory that power AI systems, and that work is dominated by TSMC. A second source in packaging would give Nvidia room to press on cost and supply at a moment when AI demand has stretched every supplier thin.
For Intel, a customer like Nvidia would mean more than a contract. The company split out its foundry as a separate business and has courted outside customers in a bid to turn around years of manufacturing stumbles. Winning a marquee name in AI chips would signal that the foundry can sell to the industry’s most demanding buyers, not just to Intel itself.
The timing matters for a second reason. Intel is in the middle of a restructuring aimed at proving it can build for others, and a public report that Nvidia is weighing Foveros — even at a distance of three years — is the kind of signal that can shape whether other customers commit.
Analysts cautioned against reading too much into a single report. Chip-design decisions made in 2028 will depend on technology road maps, pricing and politics that are all in motion now. A “may adopt” headline in 2026 has a long way to travel before it becomes a purchase order.
The deeper story is about where value in chips is migrating. As the gains from shrinking transistors slow, the companies that can assemble chips well — stacking, bonding, cooling — are the ones that capture the next round of performance. Intel built a lead in this craft and then watched others monetize the AI boom it largely missed.
Foveros is one of the assets that could change that. The technology lets designers mix chiplets made on different process nodes, cutting cost and letting a mature process carry the load for parts of a chip that do not need cutting-edge transistors. That flexibility is precisely what a company like Nvidia, selling into a market hungry for every possible unit, would want to secure.
The report, if it holds, would also be a small dent in TSMC’s standing. The Taiwanese company has been the unquestioned leader in advanced packaging for AI chips, and a second source — even one that arrives in 2028 — would chip away at that hold.
Intel’s packaging history is long, and it is the one area where the company has kept a genuine lead. Its New Mexico facility assembles chiplets at scale, and Foveros has moved from laboratory curiosity to production. The question has always been whether that capability could be sold to outsiders, and a Nvidia engagement would answer it.
None of this is decided. Intel’s foundry still has to prove it can deliver at the scale and yield Nvidia would require, and 2028 is a lifetime in semiconductors. But the fact that the report circulates at all is its own data point: the fight over AI hardware has moved past the transistor and into the packaging.
The stakes extend beyond Intel and Nvidia. Advanced packaging capacity is one of the industry’s tightest bottlenecks, and demand from AI systems has made it scarcer still. TSMC has been expanding its own packaging lines, and Samsung has pushed its own stacking technology, but supply has not kept pace with the orders flooding in. A third credible source would ease the squeeze for everyone, and Intel has been trying to position itself as exactly that.
For now, the word to watch is conditional. Nvidia “may” adopt Foveros. Intel “could” win the customer that would validate its turnaround. The market priced both possibilities the moment the report landed, the way markets always do — by betting on the story before the facts arrive.
