Gross margin went from 28% to 48% on flat volume, and the AI segment that supposedly justifies the premium contributed 1.4% of revenue.
Thesis: Powerchip is printing the cleanest pricing power signal in mature-node foundry. Flat wafer shipments plus 24% ASP plus 20-point GM expansion is textbook bottleneck pricing. The 30% input-output ASP lag mechanically guarantees Q4 margin continuation, and 80% of 2027 discrete capacity is locked under LTAs with prices already negotiated 10-20% higher. The problem is the stock's AI premium rests on a segment generating 1.4% of revenue with cleanroom capacity exhausted and new fab contribution not before 2028. The earnings are real; the growth ceiling is real. Size accordingly.
Verdict: HOLD — Conviction: MEDIUM
Catalyst: Q4 double-digit revenue growth with input ASP lag unwinding into 48%+ GM, plus new fab commitment or customer co-investment announcement (30-60% prepay model) that breaks the cleanroom ceiling.
Key Risk: Cleanroom capacity ceiling. Chairman stated flatly they 'completely cannot meet customer demand' for IPD, tools sit in warehouses, and new fab revenue contribution is 2028+. The growth beyond this pricing cycle requires capital deployment management explicitly will not commit to.
The Tell: Chairman Xie, answering why they're scouting new fab sites, said: 'to some degree, this is also comforting customers to keep hope that we'll have more capacity eventually.' He then volunteered that new capacity might come 'after the next downturn.' He admitted the fab plan is partly customer relations, not a committed expansion.
Friction Level: MODERATE_FRICTION — Both sides agree pricing power is real and capacity is full. Bull sees LTA-locked 2027 pricing as structural; bear sees the capacity ceiling as proof the AI growth narrative cannot be delivered. Same facts, opposite read on whether to own it.
2026Q3 · Oct 21, 2026
3 of 5 readings agreed3 HOLD · 1 LONG · 1 SHORTRead the full call
Pricing Power Confirmed, Growth Engine Generating 1.4% of Revenue
Powerchip is printing the cleanest pricing power signal in mature-node foundry. Flat wafer shipments plus 24% ASP plus 20-point GM expansion is textbook bottleneck pricing. The 30% input-output ASP lag mechanically guarantees Q4 margin continuation, and 80% of 2027 discrete capacity is locked under LTAs with prices already negotiated 10-20% higher. The problem is the stock's AI premium rests on a segment generating 1.4% of revenue with cleanroom capacity exhausted and new fab contribution not before 2028. The earnings are real; the growth ceiling is real. Size accordingly.
Chairman Xie, answering why they're scouting new fab sites, said: 'to some degree, this is also comforting customers to keep hope that we'll have more capacity eventually.' He then volunteered that new capacity might come 'after the next downturn.' He admitted the fab plan is partly customer relations, not a committed expansion.
Q3 beat with revenue +23% QoQ on flat volume, GM +20pts, net margin 19% to 30%. CEO guides Q4 to double-digit growth because input ASP runs 30% above shipped ASP. Memory input lag is 4 months, logic 2-3 months. The accelerator is still half pressed.
Wafer shipments declined slightly (Tongluo transition) yet ASP rose 24% and GM expanded 2,000bps. Logic prices raised 10-15% on non-LTA customers. 2027 prices already negotiated 10-20% higher under LTA. Customers demanding three-year contracts for 55nm BCD PMIC for AI servers.
80% of 2027 MOSFET/IGBT capacity locked under LTA. Logic foundry capacity fully allocated with LTA customers getting priority. TDDI lead times lengthening with panel customers showing up at the fab. PMIC customers demanding three-year binding contracts.
Cleanroom space exhausted. Tongluo tools sitting in warehouses. Chairman admits IPD industry demand is 30,000-50,000 wafers/month but they can supply 8,000-10,000. New fab in early research across Taiwan, Singapore, Japan with contribution after 2028. Chairman hedged: new capacity might arrive after the next downturn.
Management presents full utilization as bullish, but IPD production cannibalizes DRAM wafer starts on shared capacitor capacity. Every IPD wafer sold subtracts from the legacy DRAM foundry business carrying current margins. The full fab means no room for growth without displacement.
3D AI Foundry segment collapsed from 5.4% to 1.4% of revenue in one quarter. Interposer and WoW lines mid-relocation. PWF mass production Q4 2027, revenue contribution 2028. The AI-branded engine is generating almost nothing while legacy DRAM carries the quarter.
Micron funding PWF equipment in full. Chairman states new capex will be co-funded 30-60% by customers or prepaid. Depreciation dropping to NT$90-95B in 2027. Tongluo opex of NT$660M/quarter to zero by Q2 2027. Every dollar spent is customer-anchored.
-30
CAUTIOUS"CEO guided double-digit Q4 with specific ASP lag numbers, but Chairman Xie volunteered unprompted that new capacity 'might arrive after the next downturn' and that fab scouting partly exists to 'comfort customers to keep hope.' Voluntary hedging alongside a 48% GM print."
Specific ASP lag (30%), LTA coverage (80% of 2027 discrete), Tongluo opex detail (NT$660M/qtr), and depreciation guidance (NT$90-95B) all given with numbers. New fab specifics deferred entirely. Good data, clear gaps on growth capex.
Q4 double-digit revenue growth with input ASP lag unwinding into 48%+ GM, plus new fab commitment or customer co-investment announcement (30-60% prepay model) that breaks the cleanroom ceiling.
Cleanroom capacity ceiling. Chairman stated flatly they 'completely cannot meet customer demand' for IPD, tools sit in warehouses, and new fab revenue contribution is 2028+. The growth beyond this pricing cycle requires capital deployment management explicitly will not commit to.
The verdict is public. The case behind it is Pro: the forensic reading of guidance and language, the bull and bear cases, the red flags, and the play. It opens to everyone on Jan 19, 2027.