They are spending billions to qualify HBM2E in 2026 while the market moves to HBM4.
Thesis: PTI is pitching a Fan-out PLP architecture that requires customers to eat a duplicate design tax for trailing-edge HBM2E performance. Management is using 2028 revenue moonshots to mask a 300bps sequential drop in current AI revenue mix. This is not a structural win. It is a bet that hyperscalers will settle for second-best capacity in H2 2026. The restart of major capex is a high-risk gamble on a moving target.
Verdict: AVOID — Conviction: MEDIUM
Catalyst: Q2 2026 qualification results for the 2-reticle CoWoS-equivalent project and the subsequent ramp in Q3 2026.
Key Risk: TSMC CoWoS capacity remains so constrained that hyperscalers are forced to accept PTI's parallel design tax regardless of the HBM generation gap.
The Tell: Chairman Tsai admitted the 'awkward period' required customers to maintain dual designs for TSMC and PTI. He framed this as solved while revealing a qualification timeline for HBM2E that stretches into 2026. The solution arrives just as the technology becomes legacy.
Friction Level: HIGH_FRICTION — The fundamental disagreement over whether PTI's Fan-out PLP is a leapfrog architecture or a trailing-edge cost play. One side sees the end of the dual-design tax. The other sees a company qualifying HBM2E two years too late.
Report not found
The report data is no longer available. Please return to the archive.