TSMC Develops 'EMIB-like' Packaging to Block Intel's AI Chip Gains
The foundry giant is partnering with Kinsus to create a silicon-bridge architecture as CoWoS capacity constraints push customers toward Intel.
TSMC is developing a new advanced packaging technology, internally referred to as "EMIB-like," to counter the growing threat of Intel’s packaging services. The move is designed to prevent major AI customers from migrating to Intel's platform amid severe capacity shortages in TSMC's own lines.
To bring this architecture to scale, TSMC is collaborating with Kinsus Interconnect Technology Corp, a Taiwanese substrate manufacturer. The new technology aims to mimic Intel's Embedded Multi-die Interconnect Bridge (EMIB) by embedding silicon bridges directly into or closer to the package substrate. This approach provides a more flexible alternative to TSMC's standard Chip-on-Wafer-on-Substrate (CoWoS) process, which has struggled to keep pace with the explosive demand for AI hardware.
The Capacity Crisis
Advanced packaging is the critical bottleneck for modern AI chips, which require high-bandwidth connections between processors and High Bandwidth Memory (HBM). While TSMC has long dominated this space with CoWoS, the company's capacity is currently heavily constrained. This shortage has created a strategic opening for Intel, whose EMIB technology offers a more nimble approach by using localized silicon bridges rather than a full silicon interposer.
Industry reports indicate that the capacity crunch has already forced key clients to explore alternatives. Google, specifically for its 9th-generation Tensor Processing Units (TPUs), and NVIDIA have reportedly considered Intel's packaging as a viable alternative to avoid the delays associated with TSMC's choked production lines.
Strategic Implications
If TSMC successfully implements this "quasi-EMIB" solution, it could neutralize one of the few remaining competitive advantages Intel holds in the foundry market. By offering a higher-capacity, more flexible packaging option, TSMC can retain its largest AI customers and discourage them from diversifying their supply chains toward Intel's foundry services.
What to Watch
Market observers are now monitoring the timeline for the rollout of this new architecture and whether it can be scaled quickly enough to alleviate current bottlenecks. While the core strategy is clear, the full technical specifications—including whether the technology will entirely eliminate certain redistribution layers—remain to be confirmed as the project progresses with Kinsus.