ChipMOS: The EPS miss is noise, the margin trough is the investable signal

The right way to read this print is to separate what investors were set up to own from what actually broke. Priced in was a modest revenue beat, because the street was looking for 194.0 million and ChipMOS delivered 197.7 million, a +1.9% surprise that says demand was not the problem. What was not priced in was the EPS air pocket: actual EPS was -0.51 against 0.19, a -365.2% surprise. The variant perception is that the equity should not trade primarily on that EPS miss because the company’s own bridge points to foreign exchange and other nonoperating items, while the operating data show assembly utilization at 64% and average test utilization at 67%. This was not a clean quarter, but it was a quarter where the P&L looked worse than the factory.

That distinction matters because the underlying financial trajectory had already been deteriorating before the currency hit, and Q2 appears to have compressed the operating model to a level from which even normal seasonal absorption can matter. Revenue has been trapped around the mid-TWD 5 billion level since the 2024 peak, while gross margin slid to 6.6% in Q2 FY2025 from 9.4% in Q1 FY2025. Silvia Su put the company’s accounting basis plainly: “For the second quarter of 2025, total revenue was TWD 5,736 million.” That company-reported revenue sits in the same operating story as the street-comparison beat, but it should not be confused with the ADR estimate basis. The salient point is that revenue was adequate, not exceptional, and margin was the true miss.

The capacity story explains the margin damage, because this was a quarter where utilization improved but the cost base moved against ChipMOS faster than volume could absorb it. Gross profit was TWD 379 million at 6.6% gross margin, and operating expenses were TWD 424 million or 7.4% of total revenue. That is the mechanical reason operating profit narrowed to TWD 21 million and a 0.4% margin even though revenue was not weak. The company’s own explanation pins the gross-margin bridge on two quantifiable cost shocks: NTD appreciation reduced gross margin by about 1.5 ppts, while higher electricity charges reduced it by about 1.6 ppts. The summer electricity effect was not abstract inflation; it was TWD 102 million. The market may be overcapitalizing these items into the run-rate margin, when the bigger issue is whether higher utilization can outrun them in Q3.

The EPS miss becomes less informative once the nonoperating bridge is put next to the operating bridge. Silvia Su said the sequential earnings change was “mainly due to an increase of net nonoperating expenses of TWD 764 million and a decrease of operating profit of TWD 95 million,” language that matters because it allocates the damage between a large below-the-line shock and a smaller operating deterioration. The year-over-year bridge was even more explicit: foreign exchange moved from gains of TWD 25 million in Q2 2024 to losses of TWD 690 million in Q2 2025. A PM should not ignore FX in a Taiwan OSAT model, but should also not mistake a TWD 690 million FX loss for evidence that customers stopped loading the lines. The operating margin at 0.4% is bad enough to discipline the thesis; the EPS loss is not the cleanest signal.

The product mix gives the first reason to think Q2 was not simply a demand rollover. Flash represented about 29% of Q2 revenue and was up 21.7% compared to Q1, while DRAM represented 15.7% and significantly increased 19.8% compared to Q1. That is a memory-led recovery inside a company that also has meaningful display-driver exposure. The offset is that DDIC represented 23.5% of total revenue in Q2, with gold bumping about 21.2%, and the panel-related subsegments were weaker. Auto panels contributed more than 32% of Q2 DDIC revenue and were down about 13.9% compared to Q1, while OLED represented about 24.5% of Q2 DDIC revenue and decreased 14.7% compared to Q1. The signal is not broad end-market acceleration; it is mix rotation toward memory and smartphones against softer display pockets.

That rotation has second-order consequences for customers and suppliers, and the magnitudes are large enough to matter. Samsung, a ChipMOS customer for display driver IC testing and packaging, should not read this as a clean positive: TV panel demand represented 11.8% of Q2 revenue and was down 13.8% compared to Q1, while OLED DDIC was down 14.7% compared to Q1. The offset for Samsung-related display activity is that DDIC still represented 23.5% of total revenue, so the platform remains material even as the near-term mix softens. For Chang Wah Electromaterials, the supplier read-through is more balanced: gold bumping represented about 21.2% of Q2 revenue, but Gold Bump revenue decreased 7.6% compared to Q1. Memory packaging demand is the better near-term support, with Flash up 21.7% compared to Q1 and DRAM up 19.8% compared to Q1, but the driver-IC and gold-bump chain is not participating evenly.

The end-market view reinforces why the market should not price Q2 as a uniform cyclical recovery. Smartphone revenue represented 37.4% of Q2 revenue and was up 7.3% compared to Q1, which gives ChipMOS a utilization tailwind in the largest named application bucket. Consumer-related revenue represented 20.8%, while computing accounted for 4.1%, so the company is not being pulled by an AI-server compute mix in the way investors have rewarded elsewhere in semis. That matters for multiple discipline. This is a smartphone and memory absorption story with display drag, not a structural compute upgrade. The investable setup is therefore narrower: buy the trough only if memory and handset loading keep improving enough to restore gross margin, not because the company has found a new high-margin growth vector.

The comparative point is equally unforgiving. Against OSAT and packaging peers, ChipMOS’s 6.6% gross margin is far below ASX at 20.1% and 3481.TW at 14.2%, despite those peers showing revenue YoY of +17.4% and +19.2%, respectively. That gap is the opportunity and the risk. If ChipMOS can move gross margin back toward the low-teens levels already present in its own history, the stock can look mispriced on trough EPS. If it cannot, the peer comparison argues the business is absorbing volume at structurally inferior profitability. The key is that Q2 gross margin does not need heroic recovery to change the debate; it needs proof that the 6.6% level was distorted by FX, electricity, and temporary mix rather than a new steady state.

The tone of the call supports a constructive but not complacent interpretation. The tone history shows sentiment unchanged at 0.07 from Q1 FY2025 to Q2 FY2025, while guidance_tone fell by -0.19 and uncertainty rose by +4.8. That is the conflict in the transcript: management sounded no worse in broad sentiment, but the forward delivery became less assertive and more uncertain. At the same time, qa_sentiment improved by +0.11 and ai_optimism jumped by +0.91, suggesting the language model finds more forward optimism in the Q2 transcript than in Q1. I would not overread that optimism because uncertainty also moved higher, and qa_evasiveness increased by +5.9. The practical read is that management can point to utilization and mix recovery, but it did not deliver a low-uncertainty guide that would let investors underwrite a clean snapback.

That call posture fits the balance-sheet and cash-flow evidence: the company has liquidity to bridge a weak margin quarter, but not enough operating leverage yet to make the EPS miss irrelevant. Cash and cash equivalents were TWD 13,662 million as of June 30, 2025, down TWD 1,557 million compared to the beginning of the year. Net free cash inflow for the first half of 2025 was TWD 1,667 million, compared to TWD 1,433 million for the same period in 2024. The reason this matters is not that cash flow was cosmetically positive; it is that ChipMOS generated free cash inflow during the same half in which reported EPS turned negative in Q2. CapEx was also being directed across the business rather than into a single speculative bet, with Q2 split among Bumping at 20.1%, LCD driver at 31.6%, assembly at 20.1%, and testing at 28.2%. That allocation matches the mix complexity of the quarter: memory and smartphone demand improved, but DDIC capacity still consumed investment.

The market’s likely mispricing, then, is putting too much weight on the EPS headline and too little on whether Q2 utilization and mix mark the beginning of gross-margin repair. The bear case is not hard to state: gross margin of 6.6% and operating margin of 0.4% leave almost no room for error, and higher electricity alone cost TWD 102 million in Q2. The bull case is more specific than “cycle recovery”: assembly utilization at 64% and average test utilization at 67% coincided with Flash up 21.7% compared to Q1, DRAM up 19.8% compared to Q1, and smartphone revenue up 7.3% compared to Q1. If those are early loading signals rather than one-quarter restocking, gross margin should not stay at the Q2 trough.

What to watch next quarter is concrete. First, Q3 FY2025 revenue must move toward the already reported history level of TWD 6,143.7 million, because another quarter near TWD 5,735.8 million would undermine the utilization-recovery thesis. Second, gross margin needs to reclaim something close to 12.4%, the Q3 FY2025 level in the quarterly history, because remaining near 6.6% would prove that FX, power, and mix were not temporary enough. Third, listen on the next call for whether guidance_tone recovers from 0.44 and whether uncertainty falls from 21.6; the Q2 transcript had optimism but not conviction. Finally, Samsung-related DDIC and OLED commentary should stop deteriorating, while Chang Wah Electromaterials’ read-through depends on gold bumping reversing the 7.6% sequential decline. If revenue reaches TWD 6,143.7 million and gross margin approaches 12.4%, the EPS miss should be faded. If revenue stalls and gross margin stays near 6.6%, the market is right to treat Q2 as a business-model warning rather than an FX accident.