Morgan Stanley estimates Intel's EMIB-T capacity could support roughly 2.5 million TPU units by 2028 — positioning $INTC as a meaningful player in advanced packaging for AI chips. $GOOGL
Key insight: MS views Intel EMIB-T as the *primary* solution for certain TPU SKUs, with TSMC CoWoS serving as overflow capacity rather than a replacement. That's a notable shift in the packaging supply chain narrative.
MediaTek just confirmed on its earnings call that it's using Intel EMIB-T for its Google AI ASIC work — real customer validation, not just speculation.
Timeline matters: Google expected to adopt EMIB-T starting in 2027. Until then, TSMC CoWoS remains the volume workhorse across the industry through 2028.
This is about diversification and capacity relief in advanced packaging — a bottleneck that's been underestimated. Intel's foundry story isn't just about winning chip orders; it's about becoming a critical infrastructure provider in the AI supply chain.
Watch how this plays into Intel's 18A ramp and foundry credibility. If EMIB-T scales as projected, it changes the competitive dynamic in AI infrastructure beyond just logic chips.
Key insight: MS views Intel EMIB-T as the *primary* solution for certain TPU SKUs, with TSMC CoWoS serving as overflow capacity rather than a replacement. That's a notable shift in the packaging supply chain narrative.
MediaTek just confirmed on its earnings call that it's using Intel EMIB-T for its Google AI ASIC work — real customer validation, not just speculation.
Timeline matters: Google expected to adopt EMIB-T starting in 2027. Until then, TSMC CoWoS remains the volume workhorse across the industry through 2028.
This is about diversification and capacity relief in advanced packaging — a bottleneck that's been underestimated. Intel's foundry story isn't just about winning chip orders; it's about becoming a critical infrastructure provider in the AI supply chain.
Watch how this plays into Intel's 18A ramp and foundry credibility. If EMIB-T scales as projected, it changes the competitive dynamic in AI infrastructure beyond just logic chips.