Microsoft First to Adopt Intel EMIB Packaging, Cracking TSMC's Monopoly

nashnova research
今天发布阅读约 10 分钟

Microsoft will use Intel Foundry's EMIB advanced packaging for its custom Azure chips, becoming the first public adopter; this means advanced packaging — long dominated by TSMC — is now a service hyperscalers can source independently.

01

What exactly did Microsoft and Intel sign?

The two companies expanded a long-term agreement. The core commitment: Azure custom chips will use EMIB — Embedded Multi-die Interconnect Bridge, a technique that links chips from different fabs on a small silicon bridge inside the package.
The deal also locks in multi-year Xeon server CPU supply, covering the next-generation Diamond Rapids, plus processors supporting CXL interconnect, QAT acceleration, and confidential computing.
In plain terms = Microsoft signed two orders at once: one for Intel's "assembly craft" (EMIB packaging), another for Intel's "off-the-shelf chips" (Xeon CPUs).
02

What does this mean for Intel Foundry?

Intel Foundry's external business is still tiny. Q3 2026 external revenue was just $411 million, up $118 million quarter-on-quarter, mostly from one-off projects. The unit's operating loss remained $1.9 billion.
This means → the real value of Microsoft's order is not revenue — it is credibility. One of the world's largest cloud operators publicly endorsing Intel Foundry carries more weight than the dollar amount.
EMIB is also the lowest-risk entry point for any customer trying Intel Foundry — it only connects chips made at different fabs. Microsoft does not need to bet its wafer manufacturing on Intel's process roadmap.
03

Why is TSMC's packaging monopoly being pried open?

The backdrop: TSMC's CoWoS packaging capacity remains tight, with prices rising up to 20%. Analysts estimate monthly capacity will exceed 130,000 units in 2026 and could reach 200,000 in 2027 — still not enough.
Microsoft is not the only hyperscaler looking for alternatives. Per TrendForce, among three chips in Google's TPU v9 lineup, Broadcom-designed WhaleFish is expected to use TSMC CoWoS-L, while MediaTek-designed HumuFish will rely entirely on Intel EMIB-T, with volume production expected in H1 2028.
This reflects a dual-sourcing strategy — hyperscalers are not abandoning TSMC, but they are no longer putting every packaging egg in one basket.
04

Can the two packaging technologies substitute for each other?

TSMC's CoWoS-L excels at through-package redistribution-layer interconnect — rerouting chip signals across the entire packaging substrate, suited to workloads needing large-area connectivity.
Intel's EMIB excels at high-density, short-range interconnect between adjacent chips — a small silicon bridge embedded in the substrate links neighboring dies, suited to modular assembly.
In plain terms = one is built for "global wiring," the other for "dedicated express lanes between neighbors." In the near term, they serve different use cases and cannot fully replace each other.
05

Where is the real proof point?

This deal stops at the packaging layer. Microsoft has made no commitment to place wafer orders on Intel's 18A or 14A process nodes.
This means → the true inflection test for Intel Foundry is not how many packaging orders it wins, but whether packaging partnerships can eventually pull through wafer manufacturing orders.
The current state: a crack has appeared, but TSMC's double moat — advanced packaging plus wafer fabrication — has only been breached on one side.

市场有风险,内容仅供研究参考,不构成投资建议。