I’m back. My view remains unchanged: supporting two advanced-packaging solutions for the same die is both capital- and resource-intensive. For that reason, I remain skeptical that Humufish will use both CoWoS-L and EMIB-T. EMIB-T yield is improving faster than expected, with roughly another year for the process to mature. Execution still matters, but I see good reason to remain constructive. Some patience is warranted. _ MS: $TSM 's willingness to support production increases confidence in our aggressive 2028 EPS forecast: Assuming TSMC CoWoS-L can support 0.5mn TPU v9 units and EMIB-T can support 2.5mn units, our 3mn-unit target for 2028 appears highly achievable, with a higher ASP from the CoWoS version. CoWoS-L carries a structurally higher cost, and the RDL (redistribution layer) interposer connection requires tape-out of three additional photomask layers. However, our checks also suggest that Google is willing to pay these incremental costs to secure dual sourcing, which should help MediaTek sustain its gross margin.