Applied Materials and Intel have announced a research collaboration on advanced packaging for AI computing. There is no money in the release and no deadline. The interesting part is the subject: packaging is where the AI supply chain is currently stuck, and two of the largest US chip firms have just decided to work on it together.
What the two companies actually announced
Applied said it is working with Intel on next-generation transistors, interconnects and advanced packaging. The work spans Applied’s EPIC Center in Silicon Valley and Intel’s R&D campus in Hillsboro, Oregon. Applied describes Intel as a founding partner of the EPIC Center.
The verbs are all prospective. Applied “is working with” Intel, the collaboration “will bring together” capabilities, and the work “aims to shorten” development cycles. No financial terms, headcount or milestones appear.
The received version also carries a leftover placeholder. Its dateline reads “National., Oct. 8, 2026” where a city should be. Applied’s own wire release carries a Santa Clara dateline and a 6 October date.
Why advanced packaging became the bottleneck
For three years the AI industry assumed the gate on accelerator supply was logic wafer fabrication. It is not. The gate is packaging, specifically TSMC’s CoWoS lines.
TSMC’s chief executive has said CoWoS capacity remains “extremely tight and sold out through 2026”. Lead times run 52 to 78 weeks. Analysts estimate Nvidia holds roughly 60% of 2026 capacity, and the top three customers take 85% or more.
The shortage is easing at the margin. TrendForce data shows the supply-demand gap narrowing from about 20% in late 2025 to roughly 10% by mid-2026, as capacity roughly tripled since 2023. It has not closed. A fabricated 3-nanometre die without a package slot is not a sellable accelerator.
The materials problem underneath
Beneath the packaging line sits a harder constraint. A single company supplies the Ajinomoto Build-up Film used in high-performance chip substrates, and it holds roughly 95% of the global market.
Industry projections put that supply-demand gap at about 10% in late 2026, widening toward 21% in 2027 and potentially above 40% by 2028. Lead times for finished substrates run 16 to 24 weeks. Samsung Electro-Mechanics has announced a $1.2 billion expansion, but volume is not expected until the third quarter of 2027.
New capacity in substrates, interposers and packaging lines takes two to four years to build. That is why the bottleneck has proved sticky.
What Intel brings
Intel’s packaging portfolio is the reason it is a useful partner here. EMIB handles 2.5D integration through silicon bridges embedded in the substrate. Foveros handles 3D stacking, and Foveros Direct uses copper-to-copper hybrid bonding at 9-micron then 3-micron pitch.
Intel’s Rio Rancho, New Mexico site runs both, and Intel is scaling them up by 30% and 150% respectively. EMIB is already shipping for AWS, Cisco and SpaceX. Intel has confirmed that some customer designs first scoped for TSMC CoWoS were ported to Foveros without modification.
On process, Intel 18A entered production in 2025 with RibbonFET and PowerVia. Intel 18A-P entered risk production in June 2026, and Intel 18A-PT is an industry-first PowerVia-enabled base die for 3D stacking. Intel 14A is planned for risk production in 2028 and volume in 2029.
There is a caveat that matters. Intel has not landed a major external customer requiring real volume for 18A. Microsoft and the US Department of Defense are named users, but Intel itself is the main consumer. A packaging partnership does not fix that.
What Applied brings
Applied is the largest semiconductor equipment supplier, and the EPIC Center is its attempt to shorten the distance between research and volume manufacturing.
Applied announced EPIC in May 2023, with expected gross incremental investment of up to $4 billion over seven years and more than 180,000 square feet of cleanroom. It said the scale of the investment was contingent on US government support through the CHIPS and Science Act.
Intel’s founding-partner status is a label rather than a disclosed investment. When EPIC was announced, Intel was one of several companies that commented, alongside AMD, IBM, Micron, Nvidia, Samsung and TSMC. Applied has since named KIOXIA, Besi and Broadcom as EPIC partners.
The market signal
The money moving into this layer is large. SEMI puts total semiconductor equipment billings at $125 billion in 2025, forecast to pass $138 billion in 2026. Assembly and packaging equipment grew 21% in 2025 and is forecast to grow about 15% in 2026.
Worldwide 300mm fab equipment spending is expected to rise 18% to $133 billion in 2026, then 14% to $151 billion in 2027. IDTechEx forecasts advanced packaging for high-performance computing to grow at roughly 30% a year to 2037.
That is the context for a collaboration with no numbers attached. The research is early. The spending around it is not.
The counterargument
A collaboration announcement is not a product. This one commits no capital, names no customer and sets no deadline.
The scarcity that makes packaging strategic may also ease on its own. If TSMC’s Arizona packaging line qualifies on schedule, and if Amkor’s Arizona campus comes online, the overflow demand that makes Intel’s EMIB and Foveros attractive could thin by 2027.
So the honest read is narrow. Applied and Intel are aligning R&D around the industry’s real constraint, and Intel is deepening a packaging position that is already winning overflow work. The durability of that foundry story depends on customer wins Intel has not yet announced.
What to watch
Whether the collaboration produces a named deliverable, rather than a joint research programme. Whether Intel lands a volume external customer for 18A or 14A, which remains the missing piece of its foundry case. And whether substrate supply, still near-single-sourced, becomes the constraint that outlasts packaging capacity itself.
Editor’s Note
This analysis draws on the Applied Materials-Intel release of October 2026, Applied Materials’ EPIC Center announcements, Intel Foundry’s process and packaging materials, SEMI equipment forecasts, TrendForce packaging data reported in industry coverage, and published analysis of CoWoS and substrate supply. Where a figure is an estimate, we say so.