Google is testing $INTC EMIB-T. Why is it also talking to $TSM? Because if you need millions of TPUs, you want a backup before production starts, not after something goes wrong. Morgan Stanley expects Intel’s EMIB-T capacity to grow from 5k to 45k wafer equivalents per month by 2028. That’s a serious ramp. But its report also shows the risk: it estimates enough EMIB-T substrates for roughly 2 million Humufish chips, while its shipment case assumes 2.5 million from Intel plus another 0.5 million using TSMC’s CoWoS-L. That extra CoWoS-L capacity could help Google hit its target if Intel runs short. I don’t read it as Google deciding to split production between the two forever. CoWoS-L costs more, and making the same chip work with both packaging methods takes extra design work. IMO Google is keeping its options open while it learns what EMIB-T can actually deliver at scale. If Intel gets the yields and capacity right, I think it could win most of the Humufish packaging work. Intel doesn’t need to replace CoWoS across the industry for this to matter. It needs to show that Google can count on it.