TSMC CoWoS Capacity Remains Tight as Intel EMIB-T Enters Google TPU Packaging Evaluation
N.R. Finch
With TSMC's CoWoS advanced packaging capacity under sustained strain, Google is evaluating Intel's EMIB-T as an alternative packaging path for its next-gen AI TPU — a potential shake-up in the AI chip packaging supply chain.
Why is Google looking at Intel's packaging tech?
TSMC's CoWoS — an advanced packaging process that stitches multiple chips and high-bandwidth memory onto a single silicon interposer — remains capacity-constrained, with AI chip clients queuing for slots.
Google is evaluating Intel's EMIB-T (Embedded Multi-die Interconnect Bridge — a small silicon bridge linking separate chips, replacing CoWoS's full-size interposer) for its next-generation AI TPU packaging.
This means → Google is actively seeking a backup for the packaging bottleneck, rather than relying on TSMC alone.
Who else is entering Google's supply-chain evaluation?
Powerchip Semiconductor and AP Memory Technology (鈺創科技) have also entered Google's TPU supply-chain assessment.
AP Memory's silicon capacitor products reportedly play a key role in MediaTek's AI chip design project for Google.
Intel's entry into Google's view rides on the existing Google–MediaTek partnership.
What is Google itself torn about?
Google reportedly wants to hand chip design directly to TSMC for cost control, bypassing MediaTek as an intermediary.
In plain terms = Google is shopping for packaging alternatives *beyond* TSMC while simultaneously trying to cut out the middleman and go *straight to* TSMC on the design side — the two moves pull in opposite directions.
This reflects a company still weighing different supply-chain paths, with no final decision locked in.
Can EMIB actually replace CoWoS?
Intel claims EMIB offers cost and scalability advantages over CoWoS, and says performance can match SoW-class (System-on-Wafer — treating an entire wafer as one chip) levels.
But the final order allocation depends heavily on Intel's production yield — without competitive yield, the cost advantage evaporates.
This means → the technology roadmap is only the entry ticket; yield and real delivery capability are the deciding variables.
If EMIB-T wins orders, who benefits?
The most direct beneficiary is Intel itself — a marquee customer would validate its foundry-services packaging business.
The benefit could also extend to Taiwan-based packaging supply-chain players, according to reports.
In plain terms = the outcome of Google's evaluation doesn't just reshape the Intel-vs-TSMC competitive picture — it could redistribute orders across the entire packaging value chain.
Content is for reference only, not financial advice.