$SOXX up 85% since March and Nomura says we're nowhere near the top. Here's the actual alpha:
Data center projects jumped from 240 to 280 globally. Gigawatt-scale sites went 40 to 50. Compute deployment timeline now pushed past 2027.
CPU demand is massively underpriced. AMD's server CPU TAM doubled from $60B to $120B in five months. ARM targeting $100B by 2030. In agentic workloads the CPU:GPU ratio flips to 1:4, way more CPU exposure than market expects.
Real bottleneck is CoWoS packaging. TSMC targeting 2,000kpcs capacity by 2027 but Nomura models only 1,800kpcs because expansion depends on slower OSAT vendors like ASE and SPIL, not TSMC directly.
Cycle top now 2028+. Any pullback is a buy. CoWoS constraint is the sharpest edge here and consensus hasn't priced it.
Risks: TSMC historically misses its own capacity targets so 2027 output could undershoot even 1,800kpcs (tighter supply, higher pricing). Also Nomura just revised TSMC three months after last call, fast turnaround worth tracking despite solid record.
Data center projects jumped from 240 to 280 globally. Gigawatt-scale sites went 40 to 50. Compute deployment timeline now pushed past 2027.
CPU demand is massively underpriced. AMD's server CPU TAM doubled from $60B to $120B in five months. ARM targeting $100B by 2030. In agentic workloads the CPU:GPU ratio flips to 1:4, way more CPU exposure than market expects.
Real bottleneck is CoWoS packaging. TSMC targeting 2,000kpcs capacity by 2027 but Nomura models only 1,800kpcs because expansion depends on slower OSAT vendors like ASE and SPIL, not TSMC directly.
Cycle top now 2028+. Any pullback is a buy. CoWoS constraint is the sharpest edge here and consensus hasn't priced it.
Risks: TSMC historically misses its own capacity targets so 2027 output could undershoot even 1,800kpcs (tighter supply, higher pricing). Also Nomura just revised TSMC three months after last call, fast turnaround worth tracking despite solid record.