Sumco’s wafer recovery is being repriced as a depreciation cycle, not a volume cycle

Sumco Corporation did not miss because wafer demand rolled over; it disappointed because the earnings power attached to that demand is being delayed by depreciation, FX, and under-absorbed capacity. What was priced in was a return to roughly stable top line, with street revenue at ¥102,975.4 million and the company’s quarterly history already showing revenue clustered around the ¥100,000.0 million level. What actually surprised was the street-comparison print of ¥710.6 million, a -99.3% revenue surprise on that basis, while EPS was ¥0.00 against no estimate. That top-line comparison should not be blended with the company’s own accounting commentary, which discusses the quarter on a different basis, but the investment conclusion is the same either way: investors expecting a clean cyclical turn got evidence that the income statement is now dominated by fixed-cost timing rather than shipment recovery.

The company’s own numbers make that distinction sharper, because sales are no longer the main swing factor in operating profit. Shinichi Kubozoe framed the quarter with a sentence that matters because it isolates the problem: “In the analysis of sequential changes to quarterly operating profit, Q2 sales increased by JPY0.5 billion from Q1, while operating profit fell JPY4.4 billion.” In other words, the wafer market can improve modestly and Sumco can still delever operationally. The quarterly history shows revenue has been pinned near the ¥100,000.0 million band for multiple quarters, but gross margin has compressed from 31.5% at the start of the history to 17.8% in Q2 FY2025. That is not a demand-collapse chart; it is a cost-absorption chart, and the variant perception is that the next leg of earnings revisions depends less on whether AI pulls wafers and more on whether depreciation peaks before pricing and mix repair arrive.

That revenue and margin divergence is why the beat-or-miss framing is less useful than the priced-in-versus-surprised framing. The market was likely underwriting stability because Q2 FY2025 revenue in the historical series was ¥102,900.0 million and only -1.8% YoY, which looks tolerable for a wafer supplier still waiting for broader logic and memory recovery. The surprise was that Q3 guidance turns that stability into an operating loss, with company guidance for sales of ¥101 billion and an operating loss of ¥3.5 billion. A cyclical wafer trough normally invites investors to buy before utilization inflects; this print says utilization recovery is not enough if the depreciation step-up arrives first.

The depreciation detail is the core of the earnings revision risk, not an accounting footnote. Kubozoe said H1 depreciation was ¥49.4 billion, including ¥26.7 billion in Q2, and then guided Q3 depreciation to around ¥31 billion. That puts the next quarter’s margin bridge in plain view: sales are guided broadly flat, while the depreciation load rises again. The company also disclosed CapEx on an acceptance basis of ¥51.9 billion, down from ¥124.7 billion, which may eventually help free cash flow optics, but that capex deceleration does not reverse the near-term P&L burden from facilities already completed. The market may be anchoring on lower capex as a balance-sheet relief signal; the print says the income-statement drag has already been installed.

The cost story also changes how to read the AI commentary, which would otherwise sound like a clean positive catalyst. Mayuki Hashimoto said, “In terms of the outlook going forward, AI should continue to drive logic at design rules of 7-nanometer or lower and HBM within DRAM at very high levels of growth.” The wording is important because management is not saying all wafer demand is recovering; it is naming the narrow pockets that can carry the model. Those pockets map to customers such as TSMC in advanced logic, Samsung and SK Hynix in DRAM, and Micron in DRAM/NAND, but the magnitude in Sumco’s own data is still early: Hashimoto said 300-millimeter surface area growth was 9% YoY in Q2 and 6% for blended H1. That is a real demand signal, yet it is not enough to prevent an operating loss guide because the cost base is moving faster than the surface-area recovery.

That customer read-through is therefore mixed rather than generically bullish for the supply chain. For TSMC, Samsung, Intel, SK Hynix, and Micron, Sumco’s 300-millimeter data suggest advanced-node and memory wafer starts are improving, with Q2 growth of 9% YoY in the relevant surface-area metric. But for upstream suppliers such as Mersen, SGL Carbon, and Wacker Chemie, the signal is less supportive: Sumco’s CapEx on an acceptance basis fell to ¥51.9 billion from ¥124.7 billion, meaning furnace, graphite, crucible, and materials pull tied to new capacity acceptance is past its peak even as depreciation from completed projects rises. Formosa Sumco Technology, the 71% JV partner, sits between those forces: better wafer volume helps utilization, but Sumco’s operating-loss guide shows the economics of added capacity are not yet normalizing.

The balance sheet reinforces why investors should focus on timing, not just terminal demand. Total assets were ¥1,161.7 billion at the end of June, while cash and deposits were ¥80.9 billion and interest-bearing debt was ¥364.5 billion. Those figures do not imply distress, but they reduce tolerance for a long lag between volume recovery and profit recovery. Hashimoto’s comment that typical operating cash flow is between ¥70 billion to ¥80 billion and that CapEx will be less than operating cash flow gives management a cash framework, yet the equity issue is earnings duration: if Q3 sales are ¥101 billion and the operating line is -¥3.5 billion, investors are being asked to capitalize future wafer demand while current depreciation absorbs the benefit.

The peer context makes that trade-off look specific to Sumco rather than an unavoidable substrates-wide problem. In the latest peer table, SUOPY shows revenue of ¥103,264.9 million, gross margin of 6.3%, and revenue YoY of +0.8%. Comparable Japanese substrate peers in the same table show materially higher gross margins, including 25.1% for 6890.T and 32.3% for 3445.T. The point is not that those businesses have the same mix or capex cadence; the point is that Sumco’s wafer recovery is arriving with far less current margin capture. For PMs comparing substrate exposure, this print argues for a discount on Sumco’s revenue recovery until there is evidence that gross margin can move back toward the high-teens level seen before the latest depreciation step-down in profitability.

The call tone backs up that conclusion because management sounded less panicked than in Q1, but not more constructive on guidance. The tone history shows overall sentiment improved to -0.16 from -0.46, while uncertainty fell to 62.2 from 89.6. That is useful: management delivered the bad news with more clarity, not more confusion. But guidance_tone was still -0.13, and qa_sentiment was -0.28, so the better delivery does not equal a better near-term guide. The model reading is consistent with the transcript: management has a credible story for demand pockets and cost timing, but it is not yet prepared to promise that those pockets restore operating leverage next quarter.

That distinction between clarity and optimism matters because the company’s Q3 bridge is unusually explicit. Hashimoto said, “We project sales to be largely unchanged QoQ, an operating loss of JPY3.5 billion, down JPY5 billion QoQ, ordinary losses of JPY6 billion, and losses attributable to owners of the parent of JPY5.5 billion.” The commitment embedded in that sentence is not just the loss; it is the combination of flat sales and falling profit. Management also called out a negative QoQ impact of ¥0.7 billion from a stronger yen, so even if wafer shipments hold, FX adds to the depreciation drag. When a company guides flat revenue and a materially worse operating line, investors should treat volume anecdotes as insufficient unless they come with price, mix, or utilization evidence.

The longer horizon is not broken, but the sequencing is worse than bulls wanted. AI logic at 7-nanometer or lower and HBM within DRAM can raise demand for 300-millimeter wafers, and Sumco’s 9% YoY Q2 surface-area growth shows the signal is not theoretical. The problem is that the company’s nine-month projection is only ¥3.9 billion of operating profit on ¥306.3 billion of sales, with an ordinary loss of ¥1.3 billion. That means H1 profit is being mostly consumed by the Q3 reset, leaving investors with little room for another FX move, depreciation slippage, or slower non-AI wafer recovery. The market may be tempted to look through one loss quarter, but the guidance says the trough in EPS is a function of asset timing, not just customer inventories, and asset timing usually takes longer to fix.

The most actionable interpretation is to fade a simple “AI wafer recovery” rally unless the stock already reflects an operating-loss trough and a delayed margin rebuild. What was priced in was revenue stabilization around the street’s ¥102,975.4 million estimate and evidence of improving 300-millimeter demand. What actually surprised was the disconnect between that demand and operating leverage, culminating in Q3 guidance for ¥101 billion of sales and an operating loss of ¥3.5 billion. The variant perception is that Sumco is not primarily a demand-recovery call for the next quarter; it is a depreciation-absorption call, and the equity should respond more to gross margin and operating profit revisions than to AI language.

What to watch next is therefore concrete. For the next quarter, the thesis is confirmed if Sumco reports sales near the guided ¥101 billion, depreciation around ¥31 billion, and an operating loss near ¥3.5 billion, because that would prove the cost step-up is the dominant variable even with stable revenue. The thesis starts to break positively if Q3 300-millimeter surface-area growth exceeds the Q2 marker of 9% YoY and gross margin stops deteriorating from 17.8% despite the guided depreciation increase. It breaks negatively if the nine-month framework of ¥306.3 billion sales and ¥3.9 billion operating profit is cut again, because that would mean the AI and HBM pockets are not offsetting FX and fixed costs fast enough. By the next call, PMs should care less about whether management repeats the 7-nanometer and HBM language, and more about whether the operating-loss guide is the trough or the first proof that Sumco built capacity ahead of monetizable wafer demand.