EMIB-T, Substrates and Amkor: Intel’s Packaging Business Explained

When investors talk about Intel Corporation (NASDAQ: INTC) as a foundry, they usually mean wafers: 18A, 18A-P, 14A. But the part of Intel Foundry that customers are lining up for today is packaging, the step that joins several chips into one product. It is a business Intel’s CFO expects to be worth billions per customer, and it is also the reason Intel is now pre-paying billions to secure substrates.

EMIB and EMIB-T in plain words

Modern AI and server chips are too big to make as one piece of silicon. Designers split them into several dies (compute, memory, I/O) and connect them inside one package. The industry’s dominant method, TSMC’s CoWoS, places the dies on a large silicon interposer. Intel’s EMIB instead embeds small silicon bridges inside the package substrate, only where dies need to talk to each other. EMIB-T adds through-silicon vias to those bridges so power and signals can also pass vertically.

Intel’s pitch is that this removes the large interposer, allows packages well beyond the usual size limits, and costs less capital than front-end wafer manufacturing. Intel already uses EMIB in its own products, including Xeon server chips, so the technology is proven in volume.

Why customers want it, and what Intel says it is worth

Packaging capacity is one of the tightest bottlenecks in AI hardware. Intel’s management has said repeatedly that customers came to them asking for EMIB-T. At a Deutsche Bank conference in August 2026, Intel’s finance chief laid out the economics: revenue ramping in the second half of 2027, a run-rate business in 2028 and in full stride by 2029, potentially several billion dollars per customer per year, with roughly 40% gross and 30% operating margins and much lower capital intensity than wafer fabs. Intel also hired a former SK Hynix chief executive to lead the effort.

On the Q3 call Lip-Bu Tan said EMIB and EMIB-T continue to see significant external customer traction, and that Intel is adding packaging capacity in New Mexico and Malaysia for customer ramps next year. Those are forecasts and targets, not booked revenue, but they come with real money attached.

The substrate squeeze

EMIB lives inside an organic substrate, and advanced substrates are in short supply industry-wide, along with the specialised glass cloth and films that go into them. Intel’s Q3 10-Q says the shortage limited its ability to meet demand for data center CPUs and will persist into 2027. On the call Tan named substrates, alongside Intel 3 yields and the Fab 28 incident, as a headwind on server output.

Intel’s answer is money. After the quarter ended it signed two agreements to secure future substrate supply, with total funding estimated at $5.3 to $6.1 billion, of which about $3.3 billion is expected to be paid by the end of 2027. It has already made $1.2 billion of supplier prepayments this year. In May, Tan described how substrate makers in Taiwan and Japan were asking for prepayment, and how packaging customers agreed to help fund it as proof of commitment. That fits the 10-Q: in the first quarter Intel signed long-term customer arrangements that brought in $1.7 billion of deposits.

For INTC investors, this cuts both ways. The prepayments pull cash out early, in a year Intel already expects to be free-cash-flow negative. But customers putting up deposits is one of the strongest demand signals a foundry can get.

Where Amkor fits

In April 2025 Intel and Amkor announced a strategic partnership for EMIB assembly at Amkor sites in Korea, Portugal and its planned Arizona facility. A common question is how Intel earns money if Amkor does the assembly. Intel has not published the split, but the structure is familiar from the rest of the industry: Intel owns the technology and makes the silicon bridges in its own fabs, while an outsourced partner adds assembly capacity Intel cannot build fast enough itself. Outsourcing lets Intel say yes to more customers without spending its own capital on every assembly line. The trade-off is margin per package; the gain is volume and speed.

Intel is also looking further ahead. This quarter it announced a collaboration with Lens Technology on precision glass processing for future packaging, pointing toward glass-core substrates that could ease today’s organic-substrate limits.

What to watch

Three signals will show whether packaging becomes Intel’s first foundry success: named EMIB-T customers (Tan has said a couple of customer announcements are due within two quarters), the timing of payments under the substrate agreements in future 10-Qs, and external foundry revenue from the second half of 2027, when the EMIB-T ramp is supposed to begin. If those line up, packaging may prove to be the part of Intel Foundry that pays its way first.

Sources: Intel Form 10-Q for Q3 2026 (Note 14, liquidity section); Intel Q3 2026 earnings call and release; Intel at J.P. Morgan conference (May 2026) and Deutsche Bank conference (Aug 2026); Amkor, “Amkor-Intel Partnership Expands EMIB Packaging Capacity” (Apr 30, 2025).