UMC’s beat is a loading story, not a pricing story

The print changes the debate on UMC in a narrow but investable way: demand is not deteriorating, yet the incremental wafer volume is not translating into the earnings power investors usually associate with this foundry model. What was priced in was modest revenue stability and a gradual lift in fab loading after Q1; what actually surprised was the split between a +2.1% revenue beat and a -14.3% EPS miss. That separation matters more than either line item alone. The market got the top-line direction right, but missed the cost absorption problem: wafer shipments and utilization improved, while gross margin only recovered to 28.7% and diluted EPS stayed at 3.55. The actionable read is not “cyclical recovery”; it is “volume recovery with depressed conversion,” which favors customers needing stable 28nm and 40nm capacity more than it favors UMC equity until utilization and depreciation move in the same direction.

That distinction starts with the revenue base, because the top line has been pinned in a relatively tight band rather than breaking out. Management’s own basis put Q2 consolidated revenue at TWD 58.8 billion, while the street-comparison basis showed actual revenue of 1,968.6 million against 1,928.2 million. The company therefore did better than expected on demand, but not by enough to settle the earnings question. Revenue was up +1.6% sequentially and +3.4% year over year, a shape consistent with restocking or order normalization rather than a capacity-constrained upcycle. The more important historical contrast is that UMC printed gross margin above the mid-30s in 2023, while Q2 FY2025 came in at 28.7%. That is why the revenue beat did not protect EPS: the current demand level is enough to fill more tools, but not enough to restore the margin pool.

The capacity and loading data explain why the quarter feels better operationally than it looks financially. Chi-Tung Liu said wafer shipment in Q2 “increased to 967,000, up about 6.3% quarter-over-quarter,” and Jason S. Wang separately framed the Q2 wafer shipment increase as 6.2% quarter-over-quarter. The small discrepancy is not analytically important, but the direction is: shipments rose faster than revenue, which points to mix or pricing pressure rather than pure volume scarcity. Utilization moved from 69% in Q1 to 76% in Q2, yet gross margin reached only 28.7%. In a clean foundry recovery, that utilization move would be expected to carry more operating leverage. Here, the recovery is being taxed by depreciation, foreign exchange, and mix, which is why the EPS miss is the cleaner signal than the revenue beat for valuation.

The margin bridge is where the market may still be too generous. Management disclosed that first-half revenue increased by 4.7% to TWD 116 billion, while first-half gross margin fell to 27.7% from 33.1% in the same period of 2024. Those two numbers are the core of the bear case on near-term earnings: UMC is growing revenue while earning meaningfully less gross profit per unit of revenue. Liu made the depreciation issue explicit, saying that in 2023 depreciation expense increased by more than 20% year-over-year and that 2025 would see a “similar magnitude.” He also gave investors a precise FX sensitivity: every 1% NT dollar appreciation against the U.S. dollar erodes gross margin by about 0.4 to 0.5 percentage points. That is not a vague headwind. It is a direct mechanism for why revenue can beat while EPS misses, and it means the next margin inflection requires either higher utilization than the mid-70% zone or relief from currency and depreciation at the same time.

The mix data soften the downside but do not overturn it. Advanced technology below 40-nanometer reached 55% of total revenue, and 22 and 28-nanometer represented 40% of company revenue. That supports the view that UMC is not merely filling legacy capacity with low-value wafers. It also gives a specific customer read-through: MediaTek, Qualcomm, Lattice Semiconductor, and Novatek Microelectronics are tied to communication, consumer, FPGA, and OLED DDIC demand at 28nm and 40nm. Communication increased to 41%, while consumer slipped to 33%, so the Q2 mix says connectivity and communication silicon are absorbing capacity more effectively than consumer silicon. For Texas Instruments, the read-through is less about a demand surge and more about analog and specialty wafer stability, because overall UMC utilization improved to 76% but margins remained below 30%. For TSMC, the listed LSI bridge chip fabrication relationship is not large enough in this pack to make a CoWoS-L demand call, but UMC’s below-40nm share at 55% shows that specialty mature nodes remain strategically relevant alongside advanced packaging ecosystems.

That same mix is constructive for selected suppliers, but only where the numbers indicate actual fab activity rather than planned spending. The most concrete supplier signal is the utilization move from 69% to 76%, because higher wafer starts raise consumption of photomasks, resists, developers, wet chemicals, CMP consumables, and probe capacity. Photronics is the cleanest named beneficiary on photomasks, while San Fu Chemical, Everlight Chemical, Topco Scientific, and Kinik Company get a more tangible consumables read-through than equipment vendors do. Mirle Automation has a narrower angle because management said Q3 capacity increase will come mainly from 12X Xiamen, but the call did not quantify the capacity addition. The supplier implication is therefore modest: wafer-start consumables have current-quarter support, while automation and equipment are more dependent on project timing than on the Q2 shipment rebound.

The capex posture reinforces that UMC is protecting its balance sheet rather than chasing a demand spike. The company ended the first half with cash of about TWD 111 billion and first-half net income attributable to shareholders of TWD 16.67 billion. That cash cushion matters because the company is absorbing elevated depreciation while keeping capacity additions selective. Management’s strongest commitment was not to accelerate spending, but to keep the budget unchanged after the first six months. The more consequential timing detail was Jason S. Wang’s statement that “for the 12i Singapore facility, given the current market dynamics and customer alignment, we project the 12i Phase 3 production ramp will start in January 2026.” That language matters because it ties capacity to customer alignment rather than speculative demand. It also pushes a more meaningful capacity catalyst beyond the next quarter, which undercuts any thesis that Q3 alone will prove a full-margin recovery.

Against peers, UMC is neither broken nor cheap enough to ignore the margin ceiling. In the latest reported peer set, UMC showed revenue of TWD 61,037.9 million, gross margin of 29.2%, and revenue YoY of +5.5%. That puts it close to 5347.TWO on margin at 29.3% and above GFS at 27.6%, but the comparison with TSMC is the reminder of where foundry economics are diverging: TSMC printed 66.2% gross margin and +35.1% revenue YoY. The point is not that UMC should be valued like TSMC; it should not. The point is that the foundry trade is no longer a single beta call on wafer demand. UMC’s quarter says mature-node loading can improve without producing TSMC-like earnings acceleration, so relative money should prefer UMC only when the investor wants exposure to 28nm and 40nm normalization without paying for advanced-node growth.

The call delivery was better than the margin story, which creates a near-term risk of sentiment running ahead of earnings. The tone history shows Q2 FY2025 sentiment at 0.33, guidance_tone at 0.34, and ai_optimism at 0.79. Those were the highest visible readings in the series for sentiment and ai_optimism, and they coincided with lower uncertainty than Q1 FY2025, where uncertainty was 88.8 versus 72.5 in Q2. The call therefore sounded cleaner than the P&L converted. That is often tradable for one quarter, especially when revenue beats, but it is not durable unless the next print shows the same improvement in EPS. The Q&A also remained less than pristine: qa_evasiveness was 74.4 in Q2, still elevated even if below Q1 FY2025’s 89.7. The market should give management credit for clearer guidance tone, but not for earnings leverage that has not yet appeared.

That tone-to-earnings gap is the crux of the variant perception. Consensus appears to have expected a mild revenue beat or at least no demand shock, and Q2 delivered that. But the actual surprise was that EPS missed despite higher shipments and better loading. The company’s own operating facts are internally consistent: utilization improved, below-40nm mix passed half of revenue, and cash remained high. The conflict is not in the facts; it is in the valuation interpretation. If investors treat Q2 as the start of a conventional foundry recovery, they risk capitalizing revenue that is still burdened by depreciation and FX sensitivity. If they treat it as evidence that UMC’s mature-node customer base has stabilized while margins remain capped, the right position is more tactical: own dips tied to demand fears, fade rallies that assume a fast return to mid-30s gross margin.

The next quarter should be judged on three hard gates, not on the usual recovery language. First, management guided capacity utilization to the mid-70% range, so a Q3 result meaningfully above 76% would confirm that Q2 was not just shipment catch-up, while a retreat toward 69% would break the demand-stabilization thesis. Second, gross margin needs to move beyond 28.7% toward the 29.8% already shown in the quarterly history for Q3 FY2025; if it does not, depreciation and FX are still absorbing the utilization gain. Third, watch the 12X Xiamen capacity addition in Q3 and the 12i Phase 3 ramp date of January 2026, because those determine whether incremental capacity is disciplined or arrives before margins have recovered. The thesis is confirmed if Q3 keeps utilization in the mid-70% range, holds below-40nm revenue above 55%, and lifts gross margin from 28.7%. It breaks if EPS again misses while revenue beats, because that would prove the market is still mispricing UMC as a cyclical operating-leverage story rather than a mature-node stabilization story.