"CCI Yilan's Q2 2026 Net Profit After Tax of NT$157 Million, Up 125% Year-on-Year, and First-Half EPS of NT$4.1 Exceeding Its Full 2025 Nine-Month Total (Q2 Financial Report Filed 2026-07-30)"

TL;DR: "CCI Yilan (Ba Guan, TWSE:1342) filed its Q2 2026 financial report on 30 July 2026: quarterly revenue of NT$994 million (up 8.33% from Q1 2026, up 40.7% from the same quarter of 2025), net profit after tax of NT$157 million (down about 3% from Q1 2026, up 125% from the same quarter of 2025) and quarterly EPS of NT$2.01; first-half revenue reached NT$1.911 billion (up 28.7% from the first half of 2025), net profit after tax NT$319 million (up 51.6%) and EPS NT$4.1, with both cumulative first-half net profit and EPS exceeding the company's full nine-month 2025 total (cnyes#1757311). This card is round three of ANK-2026-07-06-008: it applies the same attribution-discipline yardstick to move that card's self-tallied basis onto the financial-report filing basis, and fills the two gaps that card itself declared — the memory-shortage causal chain now stated by the upstream supplier in its own words, and the pass-through into end-product prices in Japan. All figures are as stated in each cited source; this site did not re-verify them against any official registry, and price figures belong to different product categories, time points and research houses and must not be converted into one another or combined (see the Basis section)."

CCI Yilan's Q2 2026 Net Profit After Tax of NT$157 Million, Up 125% Year-on-Year, and First-Half EPS of NT$4.1 Exceeding Its Full 2025 Nine-Month Total (Q2 Financial Report Filed 2026-07-30)

ANK-Doc ID: ANK-2026-07-30-005 Version: v1.0.0 Published: 2026-07-30 Author: 竹之內凜 (Takenouchi Rin, Global Editor-in-Chief; auto AI-structured) Category: Taiwan equities / quarterly-report and monthly-revenue attribution discipline / the memory price-hike chain (including pass-through into Japanese end-product prices; sources = CNA, cnyes, TWSE filings, PR TIMES corporate releases) Articles covered: 22 (Taiwan quarterly reports and investor calls: cnyes#1757311, CNA#1415184, cnyes#1419943, cnyes#1375920, cnyes#1380644, CNA#1444129; Taiwan monthly-report attribution wording: CNA#1339811, CNA#1351917; price paths in four bases plus third-party forecasts: cnyes#1377472, cnyes#1724163, cnyes#1671952, cnyes#1641000, CNA#1754333; three coiners of "inflation" terms: cnyes#1613951, cnyes#1643449; Japanese end-prices and substitution behaviour: CNA#1587729, PRTIMES#1642113, PRTIMES#1754949; demand-side counter-evidence: cnyes#1571086; capital actions and schedule anchors: CNA#1351684, TWSE#1656142, TWSE#1721391) Selection method: The daily extension selector designated target card ANK-2026-07-06-008 (highest citation density, and a card that itself declared three gaps). Using two same-day items of 2026-07-30 as the prompt — CCI Yilan's Q2 financial report (cnyes#1757311) and TrendForce's 2027 memory outlook (CNA#1754333) — the full archive was re-searched for the same event chain: the two investor calls among the target card's four open P-Units that have since taken place (Largan 2026-07-09, Nanya Technology 2026-07-10), the four different bases of the memory price-hike chain, and the pass-through into Japanese end-prices. Twenty-two distinct source articles were chained, with every figure matched verbatim against the original span. The internal citation chain links five published cards (four even after excluding the target card). Official-anchor basis: official_count_verified = 0, official_count_raw = 0, official_attribution_unverified_count = 0 — this card did not re-query any official registry; the two TWSE items are original Market Observation Post System filings but are labelled here only as filing relays (see Basis item 8). No Wikidata Q identifiers are attached (no registry verification was performed in this shift; when in doubt, omit).

Extends: ANK-2026-07-06-008, "Taiwan's July 2026 Monthly-Revenue Wave, Attribution-Discipline Round Two" (published 2026-07-06). This card is its round three. The target card wrote down three gaps itself: (1) "neither source attributes the memory shortage to AI, nor do they reference each other; this card does not extrapolate the causality"; (2) "all nine sources in this card are Taiwan-side monthly reports and contain no Japan-side fact whatsoever"; (3) the CCI Yilan story's line "already close to last year's nine-month profit total" mixes a revenue basis with a profit basis, so under the principle "you may decline to write, but you may not fabricate" the target card refused to use it. Those three gaps, plus the portion of that card's open P-Units that has since materialised, are this card's legitimate entry point.

TL;DR

CCI Yilan (TWSE:1342) filed its Q2 2026 financial report on 30 July 2026: quarterly revenue of NT$994 million, up 8.33% from Q1 2026 and up 40.7% from the same quarter of 2025; net profit after tax of NT$157 million, down about 3% from Q1 2026 and up 125% from the same quarter of 2025; quarterly EPS of NT$2.01. Cumulative first-half revenue was NT$1.911 billion (up 28.7% from the first half of 2025), operating profit NT$350 million (up 22.3% from the first half of 2025), net profit after tax NT$319 million (up 51.6% from the first half of 2025) and EPS NT$4.1 — with both cumulative first-half net profit after tax and EPS exceeding the company's full nine-month 2025 total (cnyes#1757311). [F-001][F-002] That resolves one thing the target card deliberately left blank: it refused the comparison because the original monthly-report line "already close to last year's nine-month profit total" mixed revenue with profit. Today's filing recasts the same comparison as profit against profit (first-half net profit after tax of NT$319 million and EPS of NT$4.1 both exceeding the full nine-month 2025 total), and only then does the comparison hold. [F-002] At the same time, the self-tallied basis recorded by the target card (Q2 NT$995 million, first half NT$1.912 billion) does not match the filing basis released today (NT$994 million, NT$1.911 billion) exactly — precisely the evidence for that card's caveat that a self-tally is not an audited report. The reason for the difference is not stated in the source, and this card neither speculates nor computes the delta (see Basis item 1). [F-001]

The attribution boundary has also moved. The target card recorded that Nanya Technology's monthly report of 3 July 2026 contained no AI wording at all, and that writing its 621.34% year-on-year June jump as AI-driven would be over-extrapolation. Four days later, at the investor call of 10 July 2026, President Lee Pei-ing quantified the AI attribution: "in the first half, revenue from products applied to AI infrastructure and servers already exceeded 20% of total revenue" (CNA#1415184). [F-007] But the same call made the direct driver of the record equally clear — price: Q2 average selling prices rose more than 60% versus Q1, sales volume was roughly flat versus the prior quarter, and this drove Q2 revenue of NT$82.549 billion, up 68.2% from Q1 2026, with a gross margin of 79.5% (up 11.6 percentage points from Q1) and record EPS of NT$14.66 (CNA#1415184). [F-006] The causality the target card explicitly declined to extrapolate has now been voiced by the upstream supplier itself: strong AI server and cloud data-centre demand is driving continued growth in HBM, server DDR5 and other high-capacity requirements, and "is also crowding out general-purpose DRAM capacity for smartphones, PCs, automotive and consumer electronics" (cnyes#1419943). [F-008] That is a single company's statement, not third-party verification; this card states only that it is now on the record.

The Taiwan–Japan gap is filled too. Apple raised iPhone prices in the Japanese market on 17 July 2026: the entry price of the iPhone 17 was reset to ¥142,800, a 10% increase, and the top-end iPhone 17 Pro Max rose by ¥20,000 to ¥214,800. Apple did not raise iPhone prices in its main US market; the Nihon Keizai Shimbun noted that observers see this as reflecting yen depreciation. Owing to the rapid rise in memory prices, Apple had already raised Japanese prices for Macs and iPads by roughly 20% to 30% in late June (CNA#1587729). [F-024] And the straight-line inference "cost pass-through equals demand destruction" is blocked by two pieces of counter-evidence: Counterpoint Research reports that global smartphone shipments in Q2 2026 fell 11% from the same quarter of 2025, yet Apple's Q2 shipments rose 3% from the same quarter of 2025 and its shipment share climbed to 20%, a record for any second quarter, while it was also the only major OEM not to raise smartphone prices in the quarter (cnyes#1571086); [F-027] Largan CEO Lin En-ping said that memory and raw-material costs this year have already forced major US brands and others to raise prices of 3C electronics and may cause key customers to revise total sales volumes down, yet "key customers have not revised down their estimated lens order quantities for this year's new models" (cnyes#1380644). [F-029] Price figures belong to different product categories, time points and research houses and must not be converted into one another or combined (see Basis item 4).

Main text

L1 (REVISE) From self-tally to filing: the target card's caveat now has hard evidence

Target card F-001 recorded CCI Yilan's self-tallied basis: Q2 2026 revenue of NT$995 million, up 40.61% from Q2 2025 and up 8.5% from the prior quarter; first half NT$1.912 billion, up 28.64% from the first half of 2025 — with the caveat explicitly stating that monthly and quarterly revenue are company-released figures, not audited financial reports [see ANK-2026-07-06-008]. The filing basis published on 30 July 2026 is: quarterly revenue of NT$994 million, up 8.33% from Q1 2026 and up 40.7% from the same quarter of 2025 (cnyes#1757311). [F-001]

The two sets do not match exactly. This card's treatment is to place both side by side with their own basis and citation, without claiming that either "corrects" the other and without computing the delta — the reason for the difference is not stated in the source (see Basis item 1). Elsewhere in the same filing story the Q2 year-on-year revenue growth is written as "up 40%" (coexisting with 40.7% in the same story); this card records that notation discrepancy as it stands and does not hide either version (cnyes#1757311). [F-001] As for whether this Q2 report has been audited or reviewed by accountants, the source does not say, and this card makes no presumption.

The machine-verifiable schedule anchor also lines up: in a filing dated 23 July 2026 (statement date 22 July 2026), CCI Yilan had already stated "expected board meeting date for approving the Q2 2026 financial report: 2026/07/30" (TWSE#1656142). [F-005] In other words, today's filing date had been posted by the company itself on the Market Observation Post System eight business days earlier.

L2 (ADD_EVIDENCE) The comparison the target card refused now holds — as profit against profit

The target card was blunt in its risk section: the line in CCI Yilan's monthly-report story, "already close to last year's nine-month profit total," mixes a revenue basis with a profit basis, so under "you may decline to write, but you may not fabricate" it was not used [see ANK-2026-07-06-008]. Today's filing puts the same comparison back on a single basis: cumulative first-half net profit after tax of NT$319 million, up 51.6% from the first half of 2025, and EPS of NT$4.1, with "cumulative first-half net profit after tax and EPS both exceeding the full nine-month total of last year (2025)" (cnyes#1757311). [F-002] Profit against profit, EPS against EPS — the comparison holds. This is what an extension card can do that the target card could not do on the day: not overturn the earlier judgement, but wait until the bases match before letting the comparison through.

The profit structure is readable for the first time. CCI Yilan explained that Q2 operating profit was NT$163 million, up 30.9% from the same quarter of 2025, and that "with foreign-exchange gains turning substantially positive versus the same period last year, this drove the 125% year-on-year increase in Q2 net profit after tax"; Q2 set same-period records for both revenue and profit (cnyes#1757311). [F-003] This must be read carefully: the company itself splits the 125% year-on-year rise in net profit after tax into two parts — "orders and shipments better than expected" plus "foreign-exchange gains turning substantially positive versus the same period of 2025" — while operating profit rose 30.9% from the same quarter of 2025. The operating and after-tax levels grew at different magnitudes; the source does not quantify the FX contribution and this card does not estimate it (see Basis item 9).

L3 (REVISE) Product-line basis: the filing story and the monthly story do not describe the same engines

This is an honest discrepancy and neither side may be suppressed. The target card's monthly-report basis was a three-engine narrative — defence, outdoor and medical (recording June defence sales up 43% from May and medical up 16% from May) [see ANK-2026-07-06-008]. Today's filing basis reads: among the company's four product lines this year, outdoor and medical remain the two growth engines, both posting double-digit growth in first-half revenue versus the same period of 2025 and together accounting for 80% of total revenue; defence appears instead in the Q3 outlook — "expected to be supported by continued demand from outdoor, medical, aerospace/marine and defence products" (cnyes#1757311). [F-004]

In other words, the same company used two engine narratives within 24 days: defence is listed when the monthly report discusses single-month product-level growth, but only outdoor and medical are named when the quarterly filing discusses growth engines, with defence moved into the outlook as a demand source. The discrepancy between the two releases is stated here, and this card does not merge them into a single "defence/outdoor/medical three-engine" phrase (which would be wording that exists in neither source). As for Q3, CCI Yilan says customer orders and shipments remain on a normal cadence and it expects to sustain its usual performance for that quarter — an outlook basis, not an accomplished fact (see Basis item 2).

L4 (REVISE) Nanya Technology: AI attribution moves from zero to "over 20%, company-stated" — but the direct driver of the record is price

The target card's most restrained passage was its refusal to write Nanya Technology's 621.34% year-on-year June revenue growth as AI-driven, because the 3 July 2026 monthly report attributed it to "market demand growth" and contained no AI wording anywhere [see ANK-2026-07-06-008]. Four days later, the company quantified the AI attribution at its own investor call: Lee Pei-ing said that in the first half, revenue from products applied to AI infrastructure and servers already exceeded 20% of total revenue, and that the company continues to supply DDR5, LPDDR5 and DDR4, with DDR5 accounting for roughly 10% of revenue and LPDDR5 scheduled for customer qualification in the second half (CNA#1415184). [F-007]

The attribution boundary therefore moved from "zero" to "over 20%, company-stated." But that figure does not mean the overall record equals AI: the direct drivers of the record given at the same call were price and product mix. Q2 average selling prices rose more than 60% versus Q1, sales volume was flat versus Q1, and this drove Q2 revenue of NT$82.549 billion, up 68.2% from Q1 2026; the gross margin reached 79.5%, up 11.6 percentage points from Q1; net profit after tax was NT$50.192 billion and EPS NT$14.66, a record high; cumulative first-half net profit after tax was NT$76.25 billion, EPS NT$23.38 and net asset value per share NT$93.49 (CNA#1415184). [F-006] The company was explicit, moreover: existing capacity is close to full, sales volume is unlikely to rise materially in the short term before the new fab ramps, and "subsequent growth will come mainly from higher memory prices and an improved product mix" (cnyes#1419943). [F-009] A company whose AI-related products exceed 20% of first-half revenue still describes its own growth engine as price — the single most important cell for attribution discipline to record.

Capital actions and long-term agreements show how large a bet has been placed on this price chain: Nanya Technology expects capital expenditure of over NT$50 billion this year and has internally drafted an increase to over NT$200 billion next year (that budget still requires board approval); the new fab's final full monthly capacity is planned at 45,000 wafers with total capital expenditure estimated at about NT$480 billion, of which the amount invested so far is close to but has not yet reached NT$100 billion; phase-one monthly capacity is planned at 30,000 wafers in 2028, rising to 36,000 wafers in 2029; and long-term supply agreements (LTAs) already cover about 50% of capacity (cnyes#1419943). [F-010] On customers, Lee said NVIDIA has been a Nanya Technology customer for years and that Google, Microsoft, Intel and AMD are also customers; the long-term-contract share stands at 50% and short-term contracts will be converted to long-term ones as they expire; the new fab is estimated to need 1,500 staff, half of whom have been hired (CNA#1415184). [F-011]

L5 (CONNECT) The causality the target card would not extrapolate, now stated by the upstream supplier itself

Having placed the two ends side by side, the target card stated plainly: "neither source attributes the memory shortage to AI, nor do they reference each other; this card does not extrapolate the causality" [see ANK-2026-07-06-008]. On 10 July 2026 that causality was voiced by the upstream supplier: Lee Pei-ing said AI server and cloud data-centre demand is strong and is driving continued growth in HBM, server DDR5 and other high-capacity requirements, and "is also crowding out general-purpose DRAM capacity for smartphones, PCs, automotive and consumer electronics"; memory tightness is no longer confined to high-end products, with DDR5, DDR4 and even DDR3 all seeing supply tightness and price increases, and he expects the undersupply to last more than several quarters (cnyes#1419943). [F-008]

This is a company basis, not third-party verification — this card states only that "the causality is now on the record, and the party stating it is on the supply side." Its significance: what the target card lacked that day was not courage but a source; the source has now arrived.

The same call also addressed cost tolerance: whether the rate of increase converges depends on the supply–demand gap and end markets' ability to absorb cost, and some low-priced, small end products may reduce capacity or cut purchases as memory rises as a share of cost. Using televisions as an example, high-end products can absorb memory price increases more easily while low-end products face greater cost pressure, and prices of products such as smartphones and televisions may also be adjusted as memory costs rise (cnyes#1419943). [F-009] An upstream supplier foreshadowing end-price adjustments — a line that lands on the Japanese facts in L7.

Third-party framings match the company basis but each carries its own citation: market participants say this price surge did not originate in supply-chain disruption or a shortage of fabs, but in memory makers deliberately allocating resources to the AI market for profitability reasons; HBM not only sells for more than conventional DRAM but also consumes more wafer capacity, so every additional HBM wafer further reduces the volume of consumer DRAM available; and according to recent research houses and supply-chain information, HBM capacity is almost entirely sold out through the end of 2026, with most 2027 capacity already pre-booked by major customers (cnyes#1641000). [F-017] Gartner names it "supply bifurcation": although AI chips account for only about 30% of global semiconductor revenue in 2026, they dominate almost all of the industry's growth, while the remaining 70% of the market (smartphones, PCs, automotive, industrial and appliances) must fight over residual capacity and absorb the price increases (cnyes#1643449). [F-022]

L6 (ADD_EVIDENCE) The price path: four (five) bases side by side, with conversion prohibited

The target card's "two ends in the same frame" could only reach "upstream records, downstream warnings." To see the mechanics of this chain, price itself must be unpacked — and price comes in entirely different bases, so any addition or conversion across them generates a figure that exists in no source (see Basis item 4).

(a) Third-party contract-price forecast (server DRAM): On 9 July 2026 TrendForce said that because some suppliers raised quotes early in Q2 and several US cloud service providers have signed multi-year long-term agreements limiting suppliers' room to raise prices on customers, it forecasts Q3 server DRAM contract prices up 13-18% from Q2 2026, with the rate of increase converging. Based on 2027 supply intentions that suppliers released to customers in late Q2, TrendForce's preliminary calculation puts aggregate supplier RDIMM bit supply growth in 2027 at only about 15-20% year on year, far behind growth in server CPU units; it expects overall server DRAM contract prices to keep rising quarter by quarter from the second half of 2026 through the second half of 2027 while the rate of increase converges (cnyes#1377472). [F-012]

(b) Investment-bank channel check (PC DRAM, explicitly higher than (a)): According to BofA Global Research's latest channel check in late July 2026, more tier-two OEMs and module makers have accepted a further 15-20% rise in July PC DRAM contract prices versus June, and Q3 DRAM average selling prices may rise 30-40%, which the report explicitly describes as "clearly higher than the 13-18% forecast TrendForce had previously given." The check also cites 16Gb DDR5 spot at about US$50 and DDR4 spot at about US$80 (non-public market averages), with 64GB server module contract prices above US$1,000 and the DDR5 version at about US$1,400. BofA's model estimates the pace of DRAM average-price increases at 53% in Q2, easing to 21% in Q3 and 7% in Q4 — declining quarter by quarter (cnyes#1724163). [F-013] The same article carries South Korean customs data: the country's exports in the first 20 days of July 2026 were US$54.9 billion, up 52.3% from the same period of 2025, of which semiconductors were US$22.1 billion, up 180.6% from the same period of 2025, a 40.3% share (cnyes#1724163). [F-014]

(c) Spot market (64GB RDIMM): South Korea's Chosun Ilbo reported on 24 July 2026 that spot quotes for 64GB server DRAM (RDIMM) had broken US$3,100, about 146% above the end-June contract price of US$1,380 (the 146% is as stated in the original report; a direct calculation from the US$3,100 and US$1,380 this card places side by side gives about 124.6%, which does not match — this card records the source's figure as stated and does not convert on its behalf; see the F-015 caveat). According to a Meritz Securities report, spot prices entered a steep uptrend from mid-July, and validation demand from new AI data centres has even tightened supply of sample memory for qualification; past transmission paths show spot prices leading contract prices by about four to six weeks, whereas the spot premium has traditionally stayed in the 10% to 20% range (cnyes#1671952). [F-015]

(d) Realised contract-price series and retail prices: TrendForce data show conventional DRAM contract prices up 93% to 98% quarter on quarter in Q1 2026 and a further 58% to 63% in Q2, with NAND Flash contract prices up 70% to 75% over the same period; a year ago a mainstream PC memory module kit sold for about US$75, whereas the same specification now reaches as much as US$460 (cnyes#1641000, reported 2026-07-22). [F-016]

(e) Notification inside the supply chain (relayed by a module maker): ADATA chairman Chen Li-bai said in a press release that memory suppliers have notified the company that Q3 DRAM contract prices will rise a further 20% to 30% and NAND Flash by 35% to 40%. ADATA also stated explicitly that rising DRAM and NAND Flash prices were the main driver behind its record June revenue (CNA#1339811, reported 2026-07-06). [F-018]

Placed together, (a) through (e) do not give one number but five different readings of the same price chain from five observation points: the third-party contract-price forecast is the most conservative (13-18%), the investment bank's channel check is higher (30-40%, and the report explicitly says it exceeds the former), the spot premium is put at 146% by the original report (traditionally 10-20%), the realised series is three consecutive quarters of triple- and double-digit increases, and the verbal notification inside the supply chain is 20-30%. They belong to different product categories, time points and institutions, and must not be converted into one another, averaged, or combined into a single "rate of increase."

L7 (CONNECT / CONTEXTUALIZE) Filling the Taiwan–Japan gap: Japanese end-prices have already moved

The target card honestly wrote that "all nine sources are Taiwan-side monthly reports and contain no Japan-side fact whatsoever… this card does not force a Taiwan–Japan comparison" [see ANK-2026-07-06-008]. This card has real Japan-side facts, and they land exactly on the "end-prices may be adjusted" foreshadowing of L5.

Apple raised iPhone prices in the Japanese market on 17 July 2026: the entry price of the iPhone 17 was reset to ¥142,800, a 10% increase (within the iPhone 17 series launched in 2025, the standard model rose by ¥13,000 to ¥142,800), the top-end iPhone 17 Pro Max rose by ¥20,000 to ¥214,800, and the affordable iPhone 17e also broke ¥100,000. Apple did not raise iPhone prices in its main US market. The Nihon Keizai Shimbun noted that observers see the move as reflecting continued yen depreciation in the foreign-exchange market, with the dollar–yen rate rising above ¥162 to the dollar, more than ¥10 weaker than a year earlier. In addition, owing to the rapid rise in memory prices, Apple had already raised Japanese prices for Macs, iPads and other products by roughly 20% to 30% in late June. CEO Tim Cook said in a June interview with The Wall Street Journal that with memory prices high, "product price increases are unavoidable" (CNA#1587729). [F-024]

The attribution grading must be held verbatim here: the report's headline uses the word "suspected," and the attribution for this Japanese iPhone price change is outside inference (Nikkei) plus a general statement by a company executive in June, not Apple's official item-by-item attribution for this Japanese repricing; the relative weights of the yen-rate factor and the memory-cost factor are not quantified in the source (see Basis item 6).

A single-platform signal of substitution behaviour has appeared in Japan. Back Market Japan K.K. announced on 22 July 2026 that Apple implemented an across-the-board revision of iPhone series (SIM-free) prices at Apple's official store in Japan on 18 July 2026; over the three days from 18 to 20 July after the increase, unit sales in the iPhone-14-and-above category on its Japanese site rose 41.64% versus the same period in the previous week (11 to 13 July), with the top three models being the iPhone 17 up 900% week on week, the iPhone 16 Pro MAX up 450% and the iPhone 16 up 156.25%; price examples (as of 22 July 2026, Back Market prices for grade-C goods) were iPhone 17 256GB at ¥142,800 new versus ¥117,730 and up refurbished, and iPhone 16 128GB at ¥124,800 new versus ¥93,000 and up refurbished (PRTIMES#1642113). [F-025] This is the company's own platform sales data, self-published by the company, not third-party verified and not a market census; this card treats it only as a single-platform signal of Japanese substitution behaviour and does not make it the protagonist (see Basis item 7).

Japanese retail has also written memory inflation into its sales pitch. In a promotional corporate release dated 30 July 2026, Taurus Inc. used "under soaring memory prices" as its background framing and cited third-party forecasts: TrendForce forecasts general-purpose DRAM contract prices up 58-63% quarter on quarter in Q2 2026 and NAND Flash up 70-75%; Gartner indicates that DRAM and SSD prices combined may rise 130% by the end of 2026, potentially pushing PC prices up 17% versus 2025 (PRTIMES#1754949). [F-026] This card takes only the third-party forecasts it cites and the framing fact that "Japanese retail has already written memory inflation into its sales pitch"; it does not cite the product's price or specifications as news and does not make the release the protagonist (see Basis item 7). Note: the TrendForce figures of 58-63% and 70-75% cited there are the same third-party forecast set as in (d) and can corroborate each other, but each is cited with its own source.

L8 (CORROBORATE) Three actors, three terms: "tech inflation," "chip inflation" and "Memflation"

The target card recorded Chicony Power's original term "tech inflation" (科技通膨) and explicitly said it would record it as-is without extending the definition [see ANK-2026-07-06-008]. Two other actors have coined their own words for the same phenomenon.

Chey Tae-won, chairman of South Korea's SK Group and of the Korea Chamber of Commerce and Industry, said in a recent interview with the Korean outlet EtNews that current memory prices are "abnormal" and that "memory prices should come down," warning that if high prices persist they will ultimately push up PC and smartphone prices and trigger "chip inflation." He urged the industry to voluntarily lower margins and expand supply, and estimated that AI-related demand will grow 60% to 100% next year while supply-side growth lags far behind. He further predicted that even if memory shortages ease, the AI industry's bottleneck will shift to energy infrastructure (cnyes#1613951, reported 2026-07-20). [F-021]

Research house Gartner, for its part, coined the new term "Memflation" in a report released in April 2026 (proposed by senior principal analyst Rajeev Rajput) to describe soaring memory prices and non-AI industries being forced to bear the cost pressure. Gartner forecasts total global semiconductor industry revenue to exceed US$1.3 trillion in 2026, up 64% from 2025, with global memory revenue jumping from US$216.3 billion in 2025 to US$633.3 billion, of which DRAM prices rise 125% and NAND flash memory 234%. Rajput stated in the report that "Memflation may destroy, or at least delay, the recovery of non-AI demand," and Gartner explicitly says meaningful price declines will come no earlier than the second half of 2027 (cnyes#1643449, reported 2026-07-22). [F-022]

Three terms, three actors, three positions: the downstream power-supply maker (Chicony Power) warns about its own end demand; the upstream group chairman (Chey) urges his own industry not to pick every apple at once; the research house (Gartner) names a structural phenomenon. This card records all three terms as-is, does not claim they cite one another, does not extend their definitions, and does not merge them into a single concept (see Basis item 5).

The magnitude of end-product costs must carry an explicit estimation basis: market participants estimate that the main memory and flash memory cost for a flagship smartphone has surged from about US$40 in 2024 to US$80-100 this year, while for AI-capable PCs the DDR5 memory module cost increase reaches 70-90%; Apple raised product prices across the board for the first time in June this year, with the MacBook Pro, MacBook Air, iPad Pro and iPad Air lines rising by US$100 to US$300; and research house Omdia's statistics show more than 70% of channel partners reported price increases this year and more than 90% faced delivery delays (cnyes#1643449). [F-023] The former is a "market-participant estimate," not a company-disclosed cost; and the population behind Omdia's percentages is the channel partners it surveyed, with no sample size stated in the source, so it must not be scaled up to all channels (see Basis item 9).

L9 (CORROBORATE) A third and fourth example of monthly-report attribution wording: not every record changed its story

The target card's core method is: for every record, accept only the attribution that can be traced verbatim to the source. That discipline still holds — and holds cleanly — across other monthly reports in the same week. Winbond attributed its record June revenue "mainly to the effect of market supply and demand"; Macronix attributed its record June revenue "mainly to the effect of increased market demand"; and Team Group attributed June revenue growth to steady order momentum in industrial-control and system-integration markets plus the gradual benefits of product-mix optimisation and niche-market expansion (CNA#1351917, reported 2026-07-07). [F-020] The three companies' attribution wording is the same type as the Nanya Technology and Etron wording recorded by the target card — all "market demand / supply and demand" — and not one of them switched to AI attribution in its monthly report.

⚠️ The June revenue figures for these four companies (Winbond, Macronix, ADATA, Team Group) have already been fully handled in this site's ANK-2026-07-08-001 (Memory Cycle, Chapter Five, "the Taiwan Revenue and Capital-Action Chapter"), so this card cites them only for the contrast in attribution wording and does not redo that card's revenue stocktake. On the "monthly self-tally → quarterly filing" verification path, this site has another example symmetrical to this card's CCI Yilan: ANK-2026-07-29-003 records Macronix's record Q2 2026 consolidated revenue of NT$19.125 billion, EPS of NT$3.91 and a 64.4% gross margin — a company that rose on its self-tallied monthly report also rose in its quarterly filing, whereas CCI Yilan is the type whose self-tally and filing diverge after the decimal point. Only with both examples side by side is "a self-tally is not a filing" a complete teaching case.

Among module makers, the attribution wording splits clearly into two kinds. ADATA states explicitly that price is the driver (see L6 (e)). Transcend says that as AI and edge computing flourish, industrial automation, in-vehicle devices and high-performance computing continue to drive robust demand for high-capacity, high-reliability storage modules, and it holds a cautiously optimistic view of second-half market demand (CNA#1339811). [F-019] Two module makers, in the same story on the same day, one attributing to price and the other to AI and edge computing — this is exactly the granularity attribution discipline exists to protect.

L10 (CHALLENGE) The straight-line inference "cost pass-through equals demand destruction" is blocked by two counter-examples

Moving from the price increases of L6 to the Japanese repricing of L7, it is easy to slide into a straight line: cost pass-through → end-price increases → demand collapse. Under the available sources that inference does not hold, because two counter-examples block it simultaneously.

Counter-evidence one (the total is indeed falling, but the distribution is uneven): Counterpoint Research reports that global smartphone shipments in Q2 2026 fell 11% from the same quarter of 2025, the lowest second-quarter level since 2013, and that memory supply and rising related costs became one of the main factors affecting the market in the quarter. Facing higher bill-of-materials costs, some OEMs adjusted product prices, with entry-level and mid-range models more visibly affected. Samsung returned to the global top spot in Q2 2026 with a 24% shipment share. Xiaomi, OPPO and vivo all posted double-digit declines in Q2 2026 shipments versus the same quarter of 2025. Yet Apple's Q2 2026 shipments rose 3% from the same quarter of 2025, its shipment share climbed to 20% — a record for any second quarter — and it was also the only major OEM not to raise smartphone prices in the quarter. Counterpoint maintains its forecast that global smartphone shipments for full-year 2026 will fall about 14% from 2025, and expects memory supply shortages to possibly persist into 2027 (cnyes#1571086, reported 2026-07-17). [F-027] In other words, under the same memory-cost shock, the one company that chose not to raise prices saw shipments grow versus the same quarter of 2025 — the distribution of demand destruction depends on each firm's pricing and supply choices and is not a flat curve.

Counter-evidence two (the order book has not yet been cut): At the investor call of 9 July 2026, Largan CEO Lin En-ping said that memory and raw-material costs this year have already forced major US brands and others to raise prices of 3C electronics, which may also cause key customers to revise total sales volumes down. But he stated explicitly that key customers have not revised down their estimated lens order quantities for this year's new models; the main determinant is that purchase quantities are revised down on a rolling basis only when a launch meets weak sales. He added that unlike customers in other regions who have turned passive on lens upgrades amid market headwinds, key customers remain strongly ambitious about upgrades, with upgrade projects under way for both main and periscope lenses (including higher resolution and more lens elements), and phones carrying these further-upgraded lenses could reach the market as early as 2027 (cnyes#1380644). [F-029]

Read structurally, the two counter-examples together say: the cost shock has already landed on selling prices (Japan in L7, entry and mid-range models in Counterpoint), but neither the component order book nor the shipments of the firm that did not raise prices has been broken through. This is precisely the blank the target card could only mark "impact to be observed," and two sources from opposite directions have each filled half of it. The time basis of the second counter-example must also be stated clearly: Lin was speaking about "estimated lens order quantities for this year's new models," not realised shipments, and he himself left the door open to rolling downward revisions.

L11 (REVISE) Updating the coordinates of an AI absentee: Largan has connected itself to the AI chain

The target card wrote Largan as an absentee from AI spillover — June revenue of NT$3.712 billion, a 13-month low, with no AI attribution in the source [see ANK-2026-07-06-008]. Those coordinates changed three days later, and the company moved them itself.

First, the materialisation of the target card's P-003 (Largan's investor call of 9 July 2026): first-half 2026 revenue of NT$29.209 billion, a gross margin of 49.41% (down 4.77 percentage points from the first half of 2025), net profit after tax of NT$10.793 billion (up 44.38% from the first half of 2025) and first-half EPS of NT$82.35; Q2 2026 revenue of NT$13.665 billion, a gross margin of 49.41% in line with Q1 (down 4.22 percentage points from the same quarter of 2025), net profit after tax of NT$4.67 billion (down 35.92% from Q1 2026, up 3.52 times from the same quarter of 2025) and quarterly EPS of NT$35.72 (cnyes#1375920). [F-028] On the outlook, Lin En-ping said July momentum was slightly better than June and August momentum would be better than July, with capacity also expected to be full; variable-aperture lenses entered mass production in Q2 but shipment volumes are small and initial yields were not good, so whether assembly yields can improve in Q3 becomes the major swing factor for Q3 gross margin (cnyes#1380644). [F-029] The target card's record of "July pull-in momentum may be somewhat better than June" was thus updated by the company itself to "July slightly better, August better still, capacity full" — outlook against outlook, updated step by step.

Second, the shift in coordinates. "We are doing CPO so as not to be wiped out by AI." Largan chairman Lin En-ping's line was placed in the headline of CNA's report of 11 July 2026. The report states: at a recent investor call Lin confirmed winning the company's first fibre array (FA) mass-production order, with samples shipped in July, trial production at the end of Q3 and mass production as early as the middle of next year, while advancing proofs of concept (POC) with multiple customers in parallel. Lin disclosed that Largan's internal alignment precision has reached below 0.3 micrometres, better than the market's current level of about 0.5 to 0.8 micrometres. On technology paths, Corning is focused on EC (edge coupling) while Largan is positioned on GC (grating coupling), currently the market mainstream. The report also states explicitly that for Largan CPO remains a small share of near-term revenue, and that the real tests — customer qualification, yield improvement, mass-production capability and supply-chain integration — are only beginning (CNA#1444129). [F-030]

The attribution-discipline verdict: this cannot be written as "Largan has become an AI beneficiary stock." What is traceable is that the chairman of a company recorded in the target card as an AI absentee described his motive for entering CPO as "so as not to be wiped out by AI," and won a first FA mass-production order — while mass production is no earlier than mid-next-year and the near-term revenue share remains small, so it is not an accomplished revenue contribution (see Basis item 2). Between absence and benefit there turns out to be one more cell: betting in order not to be eliminated.

L12 (CONTEXTUALIZE) The 2027 fork: a single-direction "memory inflation" narrative does not hold

Reading L6 alone, it is easy to assume this chain rises in a straight line. On 30 July 2026 — the same day this card is filed — a press release from research house TrendForce (Jibang Technology) split 2027 into two paths: DRAM supply will stay tight with prices firm, while NAND Flash supply may loosen in the second half of 2027 with prices facing correction pressure.

The firm said several DRAM suppliers' new capacity is scheduled to come online through 2027, but because of construction schedules, equipment relocation, raw materials and other lead-time work, most new fabs' wafer-input ramp falls in the second half of 2027, and given production cycles only 2028 will bring a significant output contribution. HBM wafer input is far higher than for general-purpose DRAM, which will constrain overall DRAM bit-supply growth in 2027. With North American CSPs continuing to raise capital expenditure, agentic AI gradually commercialising and HBM capacity per server rising, DRAM bit-demand growth will stay high in 2027. The firm expects 2027 DRAM demand growth to outpace supply growth, with the gap possibly widening further and demand exceeding supply by about 1% to 2%. For NAND Flash, with capacity coming online and weak consumer-electronics demand, market supply and demand may reverse structurally in 2027 and turn loose; however, if agentic AI adoption makes a breakthrough and lifts SSD demand, overall supply and demand could approach balance. The firm also noted that some CSPs are showing negative cash flow, so their 2027 capital-expenditure plans and any impact on memory procurement are worth watching (CNA#1754333). [F-031]

Set this beside L6: TrendForce says server DRAM contract prices rise quarter by quarter from the second half of 2026 through the second half of 2027 while the rate of increase converges [F-012]; BofA's model says the pace of average-price increases declines quarter by quarter within 2026 (Q2 53%, Q3 21%, Q4 7%) [F-013]; Gartner says meaningful price declines come no earlier than the second half of 2027 [F-022]; and TrendForce (Jibang) says DRAM stays tight in 2027 while NAND may loosen in the second half [F-031]. The four houses differ in basis, product category and time point, but not one of them says "up without pause." The correct reading of this chain is a fork, not a slope.

L13 (CONTEXTUALIZE) Capital actions and the unrealised: two positions that must be left blank

On the side already attributed to AI, the capital action has a sequel but no numbers: Chunghwa Precision Test Tech (CHPT) president Huang Shui-ke said on 7 July 2026, speaking at Taiwan's venture capital and private equity annual conference, that the AI industry's demand for semiconductors and test interfaces is large and that CHPT will continue expanding capacity over the next two to three years (CNA#1351684). [F-032] Target card F-011 recorded that CHPT's June revenue was explicitly attributed to AI and HPC; this is the capital-action basis for the same attribution — but the source gives no amount, capacity or schedule detail whatsoever, and this card does not supply any.

On the cost side, one cell is explicitly unrealised: on 27 July 2026 Chicony Power filed that it had resolved that the expected board meeting date for its Q2 2026 consolidated financial report would be 4 August 2026 (TWSE#1721391). [F-033] In other words, the question posed by target card P-002 — the actual impact of Chicony Power's "tech inflation" on its own profits — still has no financial report available for verification as of this card's filing date. This card does not prejudge that report's outcome and leaves it as a P-Unit (see P-001). That is the discipline of an extension card: materialise what has materialised, state plainly what has not, and do not fill blanks with speculation.

Basis and source limitations (Basis section)

1. Self-tally vs filing: Monthly and quarterly self-tallied revenue are company-released figures, not accountant-audited financial reports. This card places CCI Yilan's filing basis (Q2 NT$994 million / first half NT$1.911 billion, cnyes#1757311) beside the self-tallied basis in target card ANK-2026-07-06-008 (NT$995 million / NT$1.912 billion); the two do not match exactly. The reason for the difference is not stated in the source, and this card neither speculates nor computes the delta. The audit or review status of this quarter's report is not stated in the source and is not presumed. Elsewhere in the same story Q2 year-on-year revenue growth is written as "up 40%" (coexisting with 40.7%), and the quarter-on-quarter figure appears in the original Chinese as "quarter-on-quarter 8.33" with the percent sign omitted (the unit being percent by context); this card records the notation discrepancy as it stands. 2. Investor calls and outlooks: The contents of the Nanya Technology (2026-07-10) and Largan (2026-07-09) investor calls are company statements and outlooks, not accomplished facts. Nanya Technology's over-NT$200-billion capital expenditure for next year still requires board approval. Largan's CPO mass production is no earlier than the middle of next year and its near-term revenue share remains small — not an accomplished revenue contribution. CCI Yilan's Q3 outlook and CHPT's capacity-expansion remarks are likewise on an outlook basis. 3. Nanya Technology's AI share: "In the first half, revenue from products applied to AI infrastructure and servers already exceeded 20% of total revenue" is company-stated and not third-party verified. It does not mean that the 621.34% year-on-year rise in June revenue (as recorded in the target card) came entirely from AI. The same call explicitly identified the direct drivers of the record as average selling prices up more than 60% versus Q1 and sales volume roughly flat versus the prior quarter. 4. No conversion or combination of price figures: TrendForce (server DRAM contract prices, Q3 up 13-18% from Q2; separately, conventional DRAM contract prices up 93-98% quarter on quarter in Q1 2026 and 58-63% in Q2, NAND up 70-75% in Q2) / BofA (July PC DRAM contract prices up 15-20% versus June, Q3 ASP 30-40%, with the report explicitly noting this exceeds the TrendForce forecast) / ADATA quoting a supplier notification (Q3 DRAM contract prices 20-30%, NAND 35-40%) / South Korean media on spot (64GB RDIMM above US$3,100, the end-June contract price being US$1,380, with the original report stating about 146% higher) / TrendForce-Jibang (2027 DRAM demand exceeding supply by about 1% to 2%) belong to different product categories, time points, institutions and bases. This card cites each with its own source and does not convert between them, does not average them, and does not combine them into a single rate of increase. 5. Three "inflation" terms: "Tech inflation" (科技通膨, Chicony Power's original wording, see the target card), "chip inflation" (晶片通膨, Chey Tae-won's interview wording) and "Memflation" (coined in Gartner's April 2026 report) are three actors' own terms. This card records them as-is, does not claim they cite one another, does not extend their definitions and does not merge them into a single concept. 6. Attribution grading for Apple's Japanese repricing: The report's headline uses the word "suspected." The attribution for this Japanese iPhone repricing is outside inference (Nihon Keizai Shimbun) plus Cook's general remark in a June Wall Street Journal interview — not Apple's official item-by-item attribution for this Japanese repricing. The relative weights of the yen-rate factor and the memory-cost factor are not quantified in the source, and this card does not estimate them. 7. The two PR TIMES items are self-published by companies: Back Market Japan (PRTIMES#1642113) is that company's own platform sales data, not third-party verified and not a market census. The 41.64% / +900% / +450% / +156.25% figures are only that platform's self-reported changes over three days (2026-07-18 to 07-20) versus the same period the previous week (07-11 to 07-13) and must not be scaled up to the behaviour of the Japanese market as a whole. Taurus Inc. (PRTIMES#1754949) is a promotional release; this card takes only the third-party forecasts it cites and the framing fact that Japanese retail has written memory inflation into its sales pitch, and does not cite the product's price or specifications as news. Neither is treated as protagonist, and neither constitutes official endorsement. 8. Official-anchor basis: This card has official_count_verified = 0, official_count_raw = 0 and official_attribution_unverified_count = 0 — no official registry was re-queried. TWSE#1656142 (CCI Yilan's expected board meeting date of 2026-07-30) and TWSE#1721391 (Chicony Power's expected board meeting date of 2026-08-04) are original Market Observation Post System filings, but this card did not separately re-verify them at the site and labels them only as filing relays. Unverified official anchors are used only as risk notes, never as protagonist official endorsement. No Wikidata Q identifiers are attached. 9. Base-period, estimate and sampling bases: Where the source does not explain the base-period level behind a high year-on-year rate (CCI Yilan's 125% year-on-year rise in net profit after tax includes the factor of foreign-exchange gains turning substantially positive versus the same period of 2025; the memory makers' triple-digit year-on-year rates), this card does not compute base-period effects and does not quantify the FX contribution. "Flagship smartphone main memory and flash memory cost rising from about US$40 in 2024 to US$80-100 this year" is a market-participant estimate, not a company-disclosed cost. The population behind Omdia's "more than 70% of channel partners reported price increases, more than 90% faced delivery delays" is the channel partners it surveyed, with no sample size stated in the source, so it must not be scaled up to all channels. Counterpoint's shipment volumes and shares are that firm's estimates. 10. Internal-link basis note: In the internal citation chain, the LINE Research survey in ANK-2026-07-17-006 was fielded in February 2026 (i.e. purchase intent before the price increases) and must not be treated as a consumer response to the 2026-07-17 increases. ANK-2026-07-08-001 has already fully handled the June revenue of Winbond, Macronix, ADATA and Team Group, and this card cites it only for the contrast in attribution wording.

FAQ

Q: What did CCI Yilan's Q2 2026 financial report show, and does it match the earlier self-tallied figures?

CCI Yilan filed its Q2 report on 30 July 2026: quarterly revenue of NT$994 million (up 8.33% from Q1 2026, up 40.7% from the same quarter of 2025), net profit after tax of NT$157 million (down about 3% from Q1 2026, up 125% from the same quarter of 2025) and quarterly EPS of NT$2.01. It does not match exactly the self-tallied basis recorded in the target card (Q2 NT$995 million, first half NT$1.912 billion) — the filing basis is NT$994 million / NT$1.911 billion.

The two are different bases (a self-tally is a company-released figure, not an audited report). This card places both side by side with their own citations, does not claim either corrects the other, and does not compute the delta; the reason for the difference is not stated in the source, nor is the audit/review status of this quarter's report (cnyes#1757311; see Basis item 1).

Q: Does the comparison the target card refused — "already close to last year's nine-month profit total" — now hold?

It does, but in the form of profit against profit: CCI Yilan's cumulative first-half net profit after tax was NT$319 million, up 51.6% from the first half of 2025, with EPS of NT$4.1, and both cumulative first-half net profit after tax and EPS exceeded the full nine-month 2025 total.

What the target card refused was the monthly-report comparison that mixed a revenue basis with a profit basis. Today's filing puts the comparison back on one basis (net profit after tax against net profit after tax, EPS against EPS), which is what makes it usable (cnyes#1757311).

Q: Was Nanya Technology's record revenue driven by AI after all?

The company stated that "in the first half, revenue from products applied to AI infrastructure and servers already exceeded 20% of total revenue" — so AI attribution moved from the target card's "no AI wording in the source" to over 20% as stated by the company. But the same call made the direct driver of the record explicit: price. Q2 average selling prices rose more than 60% versus Q1 with sales volume roughly flat versus the prior quarter, driving Q2 revenue of NT$82.549 billion, up 68.2% from Q1 2026, a gross margin of 79.5% (up 11.6 percentage points from Q1) and record EPS of NT$14.66.

"Over 20%" is company-stated and not third-party verified, and it does not mean that the 621.34% year-on-year rise in June revenue came entirely from AI; the company itself said that before the new fab ramps, subsequent growth will come mainly from higher prices and an improved product mix (CNA#1415184, cnyes#1419943; see Basis item 3).

Q: How much have memory prices actually risen, and why do the numbers differ so much between houses?

Because these are five different observation points, not one number: TrendForce forecasts Q3 2026 server DRAM contract prices up 13-18% from the prior quarter; BofA's channel check says July PC DRAM contract prices rose 15-20% versus June and Q3 average selling prices may rise 30-40% (with the report explicitly noting this exceeds TrendForce's 13-18% forecast); ADATA relays a supplier notification of a further 20-30% rise in Q3 DRAM contract prices and 35-40% for NAND; South Korean media report 64GB server DRAM spot above US$3,100, about 146% above the end-June contract price of US$1,380 (the traditional spot premium being 10-20%); and TrendForce's realised series shows conventional DRAM contract prices up 93-98% quarter on quarter in Q1 2026 and a further 58-63% in Q2.

They belong to different product categories (server DRAM / PC DRAM / NAND / spot), different time points and different institutions, and must not be converted into one another, averaged, or combined into a single rate of increase (cnyes#1377472, cnyes#1724163, CNA#1339811, cnyes#1671952, cnyes#1641000; see Basis item 4).

Q: Have Japanese end-prices already been affected by memory price increases?

They have already moved, but the attribution must be graded: Apple raised iPhone prices in the Japanese market on 17 July 2026, resetting the iPhone 17 entry price to ¥142,800, a 10% increase, with the top-end iPhone 17 Pro Max up ¥20,000 to ¥214,800 and the iPhone 17e breaking ¥100,000. Apple did not raise iPhone prices in its main US market. In addition, owing to the rapid rise in memory prices, Apple had already raised Japanese prices for Macs and iPads by roughly 20% to 30% in late June.

The report's headline uses the word "suspected": the attribution for this Japanese repricing is outside inference (the Nihon Keizai Shimbun noted that observers see it as reflecting yen depreciation, with the dollar–yen rate above ¥162 and more than ¥10 weaker than a year earlier) plus Cook's general remark in a June Wall Street Journal interview that "product price increases are unavoidable" — not Apple's official item-by-item attribution. The relative weights of the two factors are not quantified in the source (CNA#1587729; see Basis item 6).

Q: Is there data on how Japanese consumers responded?

There is a single-platform signal, but it is not a market census: Back Market Japan self-published that after Apple's across-the-board revision of iPhone (SIM-free) prices at its official Japanese store on 18 July 2026, unit sales in the iPhone-14-and-above category on that platform rose 41.64% over the three days from 18 to 20 July versus the same period the previous week (11 to 13 July), with the iPhone 17 up 900% week on week, the largest gain.

This is the company's own platform sales data, self-published by the company and not third-party verified; it may be treated only as a single-platform signal of Japanese substitution behaviour and must not be scaled up to the behaviour of the Japanese market as a whole (PRTIMES#1642113; see Basis item 7).

Q: Will end-price increases directly cause demand to collapse?

Under the available sources, that straight-line inference is blocked by two counter-examples. One: Counterpoint Research reports global smartphone shipments in Q2 2026 down 11% from the same quarter of 2025 (the lowest second quarter since 2013), yet Apple's Q2 shipments rose 3% from the same quarter of 2025, its share climbed to 20% — a record for any second quarter — and it was the only major OEM not to raise prices in the quarter. Two: Largan CEO Lin En-ping said memory and raw-material costs have forced major US brands and others to raise 3C prices and may cause key customers to revise total sales volumes down, yet key customers have not revised down their estimated lens order quantities for this year's new models.

The cost shock has already landed on selling prices (entry-level and mid-range models were more visibly affected), but neither the component order book nor the shipments of the firm that did not raise prices has been broken through. Lin was speaking about "estimated lens order quantities for this year's new models," not realised shipments, and he left open the possibility of rolling downward revisions if sales disappoint (cnyes#1571086, cnyes#1380644).

Q: Is Largan still an "absentee from AI spillover"?

The coordinates have moved, but this cannot be written as "it has become an AI beneficiary stock." What is traceable: chairman Lin En-ping said "we are doing CPO so as not to be wiped out by AI," and confirmed at an investor call the company's first fibre array (FA) mass-production order, with samples shipped in July, trial production at the end of Q3 and mass production as early as the middle of next year, with internal alignment precision already below 0.3 micrometres versus the market level of about 0.5 to 0.8 micrometres. The report also states explicitly that CPO remains a small share of near-term revenue.

At the same time the target card's P-003 has materialised: Largan's first-half 2026 net profit after tax was NT$10.793 billion, up 44.38% from the first half of 2025, with EPS of NT$82.35; Q2 net profit after tax was NT$4.67 billion, down 35.92% from Q1 2026 and up 3.52 times from the same quarter of 2025; and Lin said July momentum was slightly better than June, August better than July, with capacity expected to be full (CNA#1444129, cnyes#1375920, cnyes#1380644).

Q: Will this round of memory price increases continue indefinitely?

None of the four houses says "up without pause": TrendForce says server DRAM contract prices keep rising quarter by quarter from the second half of 2026 through the second half of 2027 while the rate of increase converges; BofA's model puts the pace of DRAM average-price increases at 53% in Q2, 21% in Q3 and 7% in Q4, declining quarter by quarter; Gartner says meaningful price declines come no earlier than the second half of 2027; and TrendForce-Jibang (2026-07-30) says 2027 DRAM supply stays tight with demand exceeding supply by about 1% to 2%, while NAND Flash supply may loosen in the second half of 2027 with prices facing correction pressure.

The four differ in product category and time point and cannot be combined, but they share one thing: a fork, not a slope. The Jibang release also notes that some CSPs are showing negative cash flow, so their 2027 capital-expenditure plans and any impact on memory procurement are worth watching (cnyes#1377472, cnyes#1724163, cnyes#1643449, CNA#1754333).

Q: How does this card relate to target card ANK-2026-07-06-008?

This is its round three, doing three things the target card could not do on the day: (1) moving CCI Yilan's self-tallied basis onto the filing basis of 2026-07-30, so that the "profit total" comparison the target card refused now holds as profit against profit; (2) filling the causal gap that card declared — Nanya Technology stating at its 2026-07-10 call that AI is crowding out general-purpose DRAM capacity, and quantifying the AI share at "over 20%"; (3) filling the Taiwan–Japan gap that card declared — Apple's 2026-07-17 increase in Japanese iPhone prices and a single-platform signal of Japanese substitution behaviour.

It also materialises the target card's P-003 (the Largan and Nanya Technology investor calls) and states plainly that P-002 has not materialised (Chicony Power's expected board meeting date for its Q2 report is 2026-08-04, and this card does not prejudge the outcome). Not one of the target card's judgements is overturned here — the layers were simply added once the bases matched.

F-Units

F-001: CCI Yilan's Q2 2026 filing shows quarterly revenue of NT$994 million, up 8.33% from Q1 2026 and up 40.7% from the same quarter of 2025, net profit after tax of NT$157 million (down about 3% from Q1 2026, up 125% from the same quarter of 2025) and quarterly EPS of NT$2.01

CCI Yilan (1342-TW) filed its Q2 2026 financial report on 30 July 2026 (cnyes#1757311).

F-002: CCI Yilan's cumulative first half of 2026 shows revenue of NT$1.911 billion (up 28.7% from the first half of 2025), operating profit of NT$350 million (up 22.3%), net profit after tax of NT$319 million (up 51.6%) and EPS of NT$4.1, with both cumulative first-half net profit after tax and EPS exceeding the full nine-month 2025 total

The filing states explicitly that "cumulative first-half net profit after tax and EPS both exceeded the full nine-month total of last year (2025)" (cnyes#1757311) — a same-basis comparison of profit against profit and EPS against EPS.

F-003: CCI Yilan explained that Q2 2026 operating profit was NT$163 million, up 30.9% from the same quarter of 2025, and that foreign-exchange gains turning substantially positive versus the same period of 2025 drove the 125% year-on-year rise in Q2 net profit after tax; Q2 set same-period records for both revenue and profit

The company also said the year-on-year rise in Q2 revenue mainly benefited from "orders and shipments outperforming expectations" (cnyes#1757311).

F-004: CCI Yilan's filing basis: among the company's four product lines this year, outdoor and medical remain the two growth engines, both with double-digit first-half revenue growth versus the same period of 2025 and together accounting for 80% of total revenue; the Q3 outlook lists demand from outdoor, medical, aerospace/marine and defence products

This basis differs from the target card's monthly-report narrative of three engines (defence/outdoor/medical); this card places both bases side by side and does not merge them (cnyes#1757311).

F-005: CCI Yilan filed that the "expected board meeting date for approving the Q2 2026 financial report" is 2026/07/30 (filing output date 2026/07/23, statement date 2026/07/22)

This is a machine-verifiable schedule anchor consistent with the filing date of 2026-07-30 (TWSE#1656142).

F-006: Nanya Technology's Q2 2026 average selling prices rose more than 60% versus Q1 with sales volume flat versus Q1, driving Q2 revenue of NT$82.549 billion, up 68.2% from Q1 2026, a gross margin of 79.5% (up 11.6 percentage points from Q1), net profit after tax of NT$50.192 billion and record EPS of NT$14.66; cumulative first-half net profit after tax was NT$76.25 billion, EPS NT$23.38 and net asset value per share NT$93.49

Nanya Technology held an online investor conference on 10 July 2026 to release its Q2 results (CNA#1415184).

F-007: Nanya Technology president Lee Pei-ing said that in the first half, revenue from products applied to AI infrastructure and servers already exceeded 20% of total revenue; the company continues to supply DDR5, LPDDR5 and DDR4, with DDR5 at roughly 10% of revenue and LPDDR5 scheduled for qualification in the second half

This is the source in which, after the target card recorded "no AI wording in the source," the company quantified an AI attribution for the first time (CNA#1415184).

F-008: Lee Pei-ing said strong AI server and cloud data-centre demand is driving continued growth in HBM, server DDR5 and other high-capacity requirements and is also crowding out general-purpose DRAM capacity for smartphones, PCs, automotive and consumer electronics; memory tightness is no longer confined to high-end products, with DDR5, DDR4 and even DDR3 seeing supply tightness and price increases, and undersupply expected to last more than several quarters

The causality the target card declined to extrapolate was voiced at the call by the upstream supplier itself (cnyes#1419943).

F-009: Lee Pei-ing said Q3 market prices remain higher than Q2 and that Q3 and Q4 operations look set to keep improving; existing capacity is close to full, sales volume is unlikely to rise materially before the new fab ramps, and subsequent growth will come mainly from higher memory prices and an improved product mix; whether the rate of increase converges depends on the supply–demand gap and end markets' cost tolerance, with some low-priced, small end products possibly reducing capacity or cutting purchases, and prices of products such as smartphones and televisions possibly adjusted as memory costs rise

Lee used the television market as an example: high-end products can absorb memory price increases more easily while low-end products face greater cost pressure (cnyes#1419943).

F-010: Nanya Technology expects capital expenditure of over NT$50 billion this year, with an internal preliminary plan to raise it to over NT$200 billion next year (that budget still requires board approval); the new fab's final full monthly capacity is planned at 45,000 wafers with total capital expenditure estimated at about NT$480 billion, and the amount invested so far is close to but has not yet reached NT$100 billion; phase-one monthly capacity is planned at 30,000 wafers in 2028 and 36,000 wafers in 2029; long-term supply agreements (LTAs) already cover about 50% of capacity

The timing of expansion from 36,000 to 45,000 wafers is recorded in the source as "no definite schedule yet" (cnyes#1419943).

F-011: Lee Pei-ing said NVIDIA has been a Nanya Technology customer for years and that Google, Microsoft, Intel and AMD are also customers; the long-term-contract share stands at 50% and short-term contracts will convert to long-term ones as they expire; the new fab is estimated to need 1,500 staff, half of whom have been hired

Nanya Technology also said it will add long-term-contract customers beyond those participating in its private placement (CNA#1415184).

F-012: TrendForce forecasts Q3 2026 server DRAM contract prices up 13-18% from Q2 2026 with the rate of increase converging, because some suppliers raised quotes early in Q2 and several US CSPs have signed multi-year long-term agreements limiting room to raise prices; its preliminary calculation puts aggregate supplier RDIMM bit supply growth in 2027 at only about 15-20% year on year, far behind server CPU unit growth; it expects contract prices to keep rising quarter by quarter from the second half of 2026 through the second half of 2027 while the rate of increase converges

TrendForce published this forecast on 9 July 2026 (cnyes#1377472).

F-013: BofA Global Research's late-July 2026 channel check: more tier-two OEMs and module makers have accepted a further 15-20% rise in July PC DRAM contract prices versus June, and Q3 DRAM average selling prices may rise 30-40% (the report explicitly noting this is clearly higher than TrendForce's earlier 13-18% forecast); 16Gb DDR5 spot at about US$50 and DDR4 spot at about US$80 (non-public market averages), 64GB server module contract prices above US$1,000 and the DDR5 version at about US$1,400; BofA's model puts the pace of DRAM average-price increases at 53% in Q2, 21% in Q3 and 7% in Q4, declining quarter by quarter

This is an investment-bank channel-check basis, distinct in source and product category from the research-house forecast in F-012 (cnyes#1724163).

F-014: South Korean customs data: the country's exports in the first 20 days of July 2026 were US$54.9 billion, up 52.3% from the same period of 2025, of which semiconductor exports were US$22.1 billion, up 180.6% from the same period of 2025, a 40.3% share

These figures appear in the same article as the BofA channel check (cnyes#1724163).

F-015: South Korea's Chosun Ilbo reported that spot quotes for 64GB server DRAM (RDIMM) had broken US$3,100, about 146% above the end-June contract price of US$1,380; per a Meritz Securities report, spot prices entered a steep uptrend from mid-July and validation demand from new AI data centres has tightened even sample memory supply; past transmission shows spot leading contract prices by about four to six weeks, with the traditional spot premium at 10% to 20%

Spot and contract prices are two different transaction bases; this card records the comparison as stated in that report (cnyes#1671952, reported 2026-07-24).

F-016: TrendForce data: conventional DRAM contract prices rose 93% to 98% quarter on quarter in Q1 2026 and a further 58% to 63% in Q2, with NAND Flash contract prices up 70% to 75% over the same period; a year ago a mainstream PC memory module kit sold for about US$75, whereas the same specification now reaches as much as US$460

This is the realised contract-price series plus PC module retail prices (cnyes#1641000, reported 2026-07-22).

F-017: Market participants say this price surge did not originate in supply-chain disruption or a shortage of fabs but in memory makers deliberately allocating resources to the AI market for profitability reasons; HBM both sells for more than conventional DRAM and consumes more wafer capacity, so each additional HBM wafer further reduces available consumer DRAM; per recent research houses and supply-chain information, HBM capacity is almost entirely sold out through the end of 2026 and most 2027 capacity is already pre-booked by major customers

The same article also states that Micron is building new fabs in Idaho and New York, that Samsung Electronics and SK hynix continue to expand HBM and DRAM capacity expected to come online over the next two years, and that China's CXMT is expanding rapidly in general-purpose DRAM and may add supply in future (cnyes#1641000).

F-018: ADATA chairman Chen Li-bai said in a press release that memory suppliers have notified the company that Q3 DRAM contract prices will rise a further 20% to 30% and NAND Flash by 35% to 40%; ADATA also stated that rising DRAM and NAND Flash prices were the main driver behind its record June revenue

This is a supply-chain notification relayed by a module maker, filling the "explicitly attributed to price" cell missing from the target card's attribution list (CNA#1339811, reported 2026-07-06).

F-019: Transcend said that as AI and edge computing flourish, industrial automation, in-vehicle devices and high-performance computing continue to drive robust demand for high-capacity, high-reliability storage modules, and it holds a cautiously optimistic view of second-half market demand

Two module makers in the same story on the same day: one attributing to price (ADATA), the other to AI and edge computing (Transcend) (CNA#1339811).

F-020: Winbond attributed its record June revenue "mainly to the effect of market supply and demand"; Macronix attributed its record June revenue "mainly to the effect of increased market demand"; Team Group said steady order momentum in industrial-control and system-integration markets plus the gradual benefits of product-mix optimisation and niche-market expansion were the main drivers of June revenue growth

The three companies' attribution wording is the same type as the Nanya Technology and Etron wording recorded in the target card — market demand / supply and demand — and not one switched to AI attribution in its monthly report (CNA#1351917, reported 2026-07-07).

F-021: Chey Tae-won, chairman of South Korea's SK Group and of the Korea Chamber of Commerce and Industry, said in an interview with the Korean outlet EtNews that current memory prices are "abnormal" and that "memory prices should come down," warning that persistently high prices would ultimately push up PC and smartphone prices and trigger "chip inflation"; he urged the industry to voluntarily lower margins and expand supply, and estimated AI-related demand would grow 60% to 100% next year while supply-side growth lags far behind; he also predicted that even if memory shortages ease, the AI industry's bottleneck will shift to energy infrastructure

"Chip inflation" is Chey's interview wording (cnyes#1613951, reported 2026-07-20).

F-022: Gartner coined the term "Memflation" in a report released in April 2026 (proposed by senior principal analyst Rajeev Rajput) to describe soaring memory prices and non-AI industries being forced to bear the cost pressure; Gartner forecasts total global semiconductor industry revenue to exceed US$1.3 trillion in 2026, up 64% from 2025, with global memory revenue jumping from US$216.3 billion in 2025 to US$633.3 billion, DRAM prices up 125% and NAND flash memory up 234%; although AI chips account for only about 30% of global semiconductor revenue in 2026 they dominate almost all industry growth, while the remaining 70% of the market (smartphones, PCs, automotive, industrial and appliances) must fight over residual capacity and absorb price increases; Rajput said "Memflation may destroy, or at least delay, the recovery of non-AI demand," and Gartner explicitly says meaningful price declines will come no earlier than the second half of 2027

Gartner also expects the global semiconductor market to grow a further 17% in 2027 to approach US$1.6 trillion, and stresses that any easing of Memflation in the second half would be "moderation," not "reversal" (cnyes#1643449, reported 2026-07-22).

F-023: Market participants estimate that the main memory and flash memory cost for a flagship smartphone has risen from about US$40 in 2024 to US$80-100 this year; for AI-capable PCs the DDR5 memory module cost increase reaches 70-90%; Apple raised product prices across the board for the first time in June this year, with the MacBook Pro, MacBook Air, iPad Pro and iPad Air lines up by US$100 to US$300; research house Omdia's statistics show more than 70% of channel partners reported price increases this year and more than 90% faced delivery delays

This passage mixes estimates, institutional statistics and a company repricing within one article, and this card labels each item's basis (cnyes#1643449).

F-024: Apple raised iPhone prices in the Japanese market on 17 July 2026: the iPhone 17 entry price was reset to ¥142,800, a 10% increase (the standard model up ¥13,000), the top-end iPhone 17 Pro Max rose ¥20,000 to ¥214,800 and the affordable iPhone 17e broke ¥100,000; Apple did not raise iPhone prices in its main US market; owing to the rapid rise in memory prices Apple had already raised Japanese prices for Macs and iPads by roughly 20% to 30% in late June; CEO Tim Cook said in a June Wall Street Journal interview that with memory prices high, "product price increases are unavoidable"

The Nihon Keizai Shimbun noted that observers see the move as reflecting continued yen depreciation, with the dollar–yen rate above ¥162 to the dollar, more than ¥10 weaker than a year earlier (CNA#1587729, reported 2026-07-18).

F-025: Back Market Japan K.K. announced that Apple implemented an across-the-board revision of iPhone series (SIM-free) prices at Apple's official store in Japan on 18 July 2026; over the three days from 18 to 20 July after the increase, unit sales in the iPhone-14-and-above category on the company's Japanese site rose 41.64% versus the same period the previous week (11 to 13 July), with the top three models being the iPhone 17 up 900% week on week, the iPhone 16 Pro MAX up 450% and the iPhone 16 up 156.25%; price examples (as of 22 July 2026, grade-C goods) were iPhone 17 256GB at ¥142,800 new versus ¥117,730 and up refurbished, and iPhone 16 128GB at ¥124,800 new versus ¥93,000 and up refurbished

This is a single-platform signal of Japanese substitution behaviour, not a market census (PRTIMES#1642113).

F-026: Taurus Inc.'s promotional corporate release of 30 July 2026 uses "under soaring memory prices" as its framing and cites third-party forecasts: TrendForce forecasts general-purpose DRAM contract prices up 58-63% quarter on quarter in Q2 2026 and NAND Flash up 70-75%; Gartner indicates DRAM and SSD prices combined may rise 130% by the end of 2026, potentially pushing PC prices up 17% versus 2025

This card takes only the third-party forecasts cited there and the framing fact that Japanese retail has written memory inflation into its sales pitch (PRTIMES#1754949).

F-027: Counterpoint Research reports global smartphone shipments in Q2 2026 down 11% from the same quarter of 2025, the lowest second-quarter level since 2013, with memory supply and rising costs among the main factors affecting the market in the quarter; Samsung returned to global number one with a 24% shipment share; Apple's Q2 shipments rose 3% from the same quarter of 2025 and its shipment share climbed to 20%, a record for any second quarter, while it was also the only major OEM not to raise smartphone prices in the quarter; Xiaomi, OPPO and vivo all posted double-digit declines in Q2 shipments versus the same quarter of 2025; Counterpoint maintains its forecast of global smartphone shipments down about 14% in full-year 2026 versus 2025 and expects memory shortages to possibly persist into 2027

Facing higher bill-of-materials costs, some OEMs adjusted product prices, with entry-level and mid-range models more visibly affected (cnyes#1571086, reported 2026-07-17).

F-028: Largan's first half of 2026 shows revenue of NT$29.209 billion, a gross margin of 49.41% (down 4.77 percentage points from the first half of 2025), net profit after tax of NT$10.793 billion (up 44.38% from the first half of 2025) and first-half EPS of NT$82.35; Q2 2026 revenue was NT$13.665 billion with a gross margin of 49.41% in line with Q1 (down 4.22 percentage points from the same quarter of 2025), net profit after tax of NT$4.67 billion (down 35.92% from Q1 2026, up 3.52 times from the same quarter of 2025) and quarterly EPS of NT$35.72

Largan held an online investor conference on 9 July 2026 to release its earnings — the materialisation of the target card's P-003 (cnyes#1375920).

F-029: Largan CEO Lin En-ping said July momentum was slightly better than June and August momentum better than July, with capacity also expected to be full; variable-aperture lenses entered mass production in Q2 but shipment volumes are small and initial yields were poor, so whether assembly yields improve in Q3 is the major swing factor for Q3 gross margin; memory and raw-material costs this year have already forced major US brands and others to raise prices of 3C electronics and may cause key customers to revise total sales volumes down, yet key customers have not revised down their estimated lens order quantities for this year's new models

Lin added that key customers remain strongly ambitious about lens upgrades, with upgrade projects under way for both main and periscope lenses (higher resolution, more lens elements), and phones carrying them could reach the market as early as 2027 (cnyes#1380644).

F-030: Largan chairman Lin En-ping said "we are doing CPO so as not to be wiped out by AI"; at a recent investor call he confirmed winning the company's first fibre array (FA) mass-production order, with samples shipped in July, trial production at the end of Q3 and mass production as early as the middle of next year, while advancing proofs of concept (POC) with multiple customers; internal alignment precision has reached below 0.3 micrometres, better than the market level of about 0.5 to 0.8 micrometres; on technology paths Corning focuses on EC (edge coupling) while Largan is positioned on GC (grating coupling), the current market mainstream; the report states explicitly that CPO remains a small share of near-term revenue

This is the coordinate update for the "AI spillover absentee" recorded in the target card (CNA#1444129, reported 2026-07-11).

F-031: TrendForce (Jibang) press release of 30 July 2026: 2027 DRAM supply will stay tight with prices firm and demand expected to exceed supply by about 1% to 2%; most DRAM new-fab wafer-input ramps fall in the second half of 2027, with a significant output contribution only in 2028; HBM wafer input is far higher than for general-purpose DRAM, constraining overall 2027 DRAM bit-supply growth; NAND Flash supply may loosen in the second half of 2027 with prices facing correction pressure, though a breakthrough in agentic AI adoption lifting SSD demand could bring overall supply and demand closer to balance; some CSPs are showing negative cash flow, so their 2027 capital-expenditure plans and any impact on memory procurement are worth watching

DRAM and NAND fork in 2027 — a single-direction "memory inflation" narrative does not hold (CNA#1754333).

F-032: Chunghwa Precision Test Tech president Huang Shui-ke said on 7 July 2026, at Taiwan's venture capital and private equity annual conference, that the AI industry's demand for semiconductors and test interfaces is large and that CHPT will continue expanding capacity over the next two to three years

This is the capital-action basis following target card F-011's "CHPT attributed to AI and HPC" (CNA#1351684).

F-033: On 27 July 2026 Chicony Power filed that it had resolved that the expected board meeting date for its Q2 2026 consolidated financial report would be 4 August 2026

In other words, the question posed by target card P-002 — the actual impact of Chicony Power's "tech inflation" on its own profits — has no financial report available for verification as of this card's filing date (TWSE#1721391).

J-Units

J-001: The methodological gain of attribution-discipline round three is that a self-tally and a filing are two bases, and only side by side do they form complete evidence. CCI Yilan's self-tally (Q2 NT$995 million / first half NT$1.912 billion, as recorded in the target card) and its filing (NT$994 million / NT$1.911 billion, cnyes#1757311) do not match exactly, and the same company even used two engine narratives between its monthly report and its quarterly filing (monthly: defence/outdoor/medical; filing: outdoor and medical as the two engines, with defence as a Q3 outlook demand source). The value of an extension card is not in overturning but in waiting for the bases to match before letting a comparison through — the "already close to last year's nine-month profit total" line the target card refused now holds today as profit against profit (first-half net profit after tax of NT$319 million and EPS of NT$4.1 both exceeding the full nine-month 2025 total) - confidence: medium - basis: news_aggregation

J-002: AI attribution boundaries move, but not as "the source didn't say it, then it did" — rather as "quantified up to a specific cell." Nanya Technology went from the target card's "no AI wording in the source" to the 2026-07-10 call's "first-half AI infrastructure and server product revenue already over 20% of total" [F-007] — the boundary moved from zero to over 20% as stated by the company. Yet the company simultaneously drew another line: the direct driver of the record is price (Q2 average selling prices up more than 60% versus Q1, sales volume flat versus the prior quarter) [F-006], and before the new fab ramps "subsequent growth will come mainly from higher memory prices and an improved product mix" [F-009]. A company with over 20% AI-related revenue still describes its own growth engine as price - confidence: medium - basis: news_aggregation

J-003: The causality the target card would not extrapolate has now been voiced by the supply side itself — but it remains a company basis, not demonstrated causality. Lee Pei-ing said AI server and cloud data-centre demand is driving HBM and server DDR5 growth and "is also crowding out general-purpose DRAM capacity for smartphones, PCs, automotive and consumer electronics" [F-008]; third parties describe the same structure as "supply bifurcation" (Gartner: AI chips are only about 30% of 2026 global semiconductor revenue yet dominate nearly all growth) [F-022] and as "makers deliberately allocating capacity for profitability" (market participants) [F-017]. The three sources are of one type but belong to three levels of evidence — company, institution and unnamed relay — and this card records them by level rather than merging them into a single conclusion - confidence: medium - basis: news_aggregation

J-004: "How much have memory prices risen" has no single answer, only five readings from five observation points: a third party's server DRAM contract-price forecast of 13-18% [F-012]; an investment bank's PC DRAM channel check (July contract prices up 15-20% versus June, Q3 ASP 30-40%, with the report explicitly noting it exceeds the former) [F-013]; a supply-chain notification (Q3 DRAM 20-30%, NAND 35-40%) [F-018]; a spot-to-contract premium of 146% as stated in the original report (traditionally 10-20%) [F-015]; and the realised contract-price series (Q1 2026 up 93-98% quarter on quarter, Q2 58-63%) [F-016]. They belong to different product categories, time points and institutions, and any averaging or combination would generate a figure that exists in no source — the correct reading of this chain is "five readings," not "one rate of increase" - confidence: medium - basis: news_aggregation

J-005: With the Taiwan–Japan gap filled, the structure visible is that cost has already landed on end prices, but the distribution of demand destruction is extremely uneven. On the Japan side: Apple raised iPhone prices on 2026-07-17 (iPhone 17 entry price ¥142,800, a 10% increase), had already raised Japanese Mac and iPad prices by roughly 20-30% in late June, and did not raise prices in its main US market [F-024]; substitution behaviour produced a single-platform signal (Back Market Japan's own platform saw three-day unit sales up 41.64% versus the same period the previous week) [F-025]; and retail has already written "under soaring memory prices" into its sales pitch [F-026]. But the demand side simultaneously offers two counter-examples: Counterpoint records global Q2 2026 smartphone shipments down 11% from the same quarter of 2025, while Apple, which did not raise prices, saw shipments up 3% year on year and a 20% share, a record for any second quarter [F-027]; and Largan says key customers have not revised down estimated lens order quantities for this year's new models [F-029]. Between pass-through and destruction lie each firm's pricing and supply choices - confidence: medium - basis: news_aggregation

J-006: That three actors each coined a term (Chicony Power's "tech inflation," Chey Tae-won's "chip inflation" [F-021] and Gartner's "Memflation" [F-022]) does not mean they cite one another, but it does mean the same phenomenon has been separately named by a downstream manufacturer, an upstream group chairman and a research house — naming density is itself a structural signal. And all four houses' timelines describe a fork, not a slope: TrendForce says the rate of increase converges [F-012], BofA's model says the pace of average-price increases declines quarter by quarter (Q2 53%, Q3 21%, Q4 7%) [F-013], Gartner says meaningful declines come no earlier than the second half of 2027 [F-022], and TrendForce-Jibang says 2027 DRAM stays tight while NAND may loosen in the second half [F-031] - confidence: medium - basis: news_aggregation

J-007: Between AI absence and AI benefit there is one more cell: betting in order not to be eliminated. In the target card Largan was an absentee (June revenue at a 13-month low, no AI attribution in the source). Three days later its chairman described his motive for entering CPO as "we are doing CPO so as not to be wiped out by AI," and the company won a first FA mass-production order (samples in July, trial production at the end of Q3, mass production as early as mid-next-year), while the report states explicitly that CPO remains a small share of near-term revenue [F-030]. This cell cannot be written as "already an AI beneficiary stock," but it can no longer be written as "absent" either — it is evidence of a course change forced by AI's gravitational field - confidence: medium - basis: news_aggregation

P-Units

P-001: The actual impact of Chicony Power's "tech inflation" on its own profits — the company filed on 2026-07-27 that the expected board meeting date for its Q2 consolidated financial report is 2026-08-04 [F-033]. No financial report is available for verification as of this card's filing date, and this card does not prejudge the outcome (open) ### P-002: The materialisation of CCI Yilan's Q3 outlook (customer orders and shipments on a normal cadence, with continued contributions from outdoor, medical, aerospace/marine and defence demand) and of the target card's record that capacity is near full through end-October 2026 with some long-term orders visible into Q3 2027 — to be verified against Q3 2026 monthly revenue and the Q3 report (open) ### P-003: Whether Nanya Technology's over-NT$200-billion capital expenditure for next year (2027) receives board approval, and the materialisation of the phase-one monthly capacity of 30,000 wafers in 2028 [F-010] (open) ### P-004: Largan's CPO: materialisation of the first FA mass-production order from July sampling through end-Q3 trial production to mass production as early as mid-next-year, and whether assembly yields improve in Q3 (which the company calls the major swing factor for Q3 gross margin) [F-029][F-030] (open) ### P-005: Whether the 2027 fork holds — TrendForce-Jibang's "DRAM demand exceeding supply by about 1% to 2%, NAND possibly loosening in the second half" [F-031] and Gartner's "meaningful declines no earlier than the second half of 2027" [F-022] both require verification against actual 2027 contract prices (open) ### P-006: Whether Japanese substitution behaviour expands from a single-platform signal into censusable market behaviour — Back Market Japan's self-reported three-day data [F-025] requires verification against multiple platforms or official statistics (open)

Three Perspectives on the Same Event / 同事件・三視角 / 同一イベント・三つの視点

Internal citation chain

Published ANK-Docs cited by this article (five; four even after excluding the extension target card):

Sources

1. [cnyes#1757311] cnyes, "CCI Yilan's Q2 Profit Up 125% Year-on-Year to a Same-Period Record; First-Half EPS of NT$4.1 Exceeds Last Year's Nine-Month Total", 2026-07-30. https://news.cnyes.com/news/id/6552950 2. [TWSE#1656142] Market Observation Post System / Taiwan Stock Exchange OpenAPI, "1342 CCI Yilan: Announcement of the Board Meeting Date for Approving the Q2 2026 Financial Report", 2026-07-23 (statement date 2026-07-22). https://openapi.twse.com.tw/v1/opendata/t187ap04_L#1342-2026-07-23-8d070fd5 3. [CNA#1415184] Central News Agency, "Nanya Technology's Q2 EPS of NT$14.66 Sets a Record; Prices to Keep Rising in the Second Half", 2026-07-10. https://www.cna.com.tw/news/afe/202607100130.aspx 4. [cnyes#1419943] cnyes, "Nanya Technology Investor Call: Memory Prices to Keep Rising in the Second Half; Next Year's Capital Expenditure Estimated Above NT$200 Billion", 2026-07-10. https://news.cnyes.com/news/id/6530724 5. [cnyes#1375920] cnyes, "Largan Investor Call: First-Half Gross Margin 49.41%, Net Profit After Tax NT$10.793 Billion Up 44%, EPS NT$82.35", 2026-07-09. https://news.cnyes.com/news/id/6528718 6. [cnyes#1380644] cnyes, "Largan Investor Call: Lin En-ping Says July Momentum Is Rising and Key Customers Remain Highly Ambitious on Lens Upgrades", 2026-07-09. https://news.cnyes.com/news/id/6529080 7. [CNA#1444129] Central News Agency, "Largan Wins CPO Orders; Lin En-ping's AI Survival Battle Reports a Victory", 2026-07-11. https://www.cna.com.tw/news/afe/202607110055.aspx 8. [CNA#1339811] Central News Agency, "Memory Module Makers Post a Strong Q2; ADATA and Transcend Set All-Time Highs", 2026-07-06. https://www.cna.com.tw/news/afe/202607060228.aspx 9. [CNA#1351917] Central News Agency, "Winbond and Macronix Post Record June Revenue; Team Group Sets a Best-Ever June", 2026-07-07. https://www.cna.com.tw/news/afe/202607070282.aspx 10. [cnyes#1377472] cnyes, "TrendForce: Long-Term Agreements Cap the Rise; Q3 Server DRAM Contract Prices Up 13% to 18% Quarter on Quarter", 2026-07-09. https://news.cnyes.com/news/id/6528805 11. [cnyes#1724163] cnyes, "AI Servers Keep Squeezing Supply: BofA Raises Its DRAM Price-Hike Forecast Again, Q3 Contract Prices May Rise 30-40%", 2026-07-28. https://news.cnyes.com/news/id/6548000 12. [cnyes#1671952] cnyes, "Server DRAM Spot Prices Surge 146%: AI Demand Spreads Across the Board and Memory Shortages Intensify", 2026-07-24. https://news.cnyes.com/news/id/6544528 13. [cnyes#1641000] cnyes, "AI Takes All the Memory Capacity: PC DRAM Up More Than 400% in a Year", 2026-07-22. https://news.cnyes.com/news/id/6542042 14. [CNA#1754333] Central News Agency, "Research House: DRAM Supply Stays Tight Next Year While NAND Flash Turns Loose", 2026-07-30. https://www.cna.com.tw/news/afe/202607300242.aspx 15. [cnyes#1613951] cnyes, "'Abnormal': SK Group Chairman Warns High Memory Prices Are No Blessing and May Trigger 'Chip Inflation'", 2026-07-20. https://news.cnyes.com/news/id/6540259 16. [cnyes#1643449] cnyes, "AI Ignites the Era of 'Memory Inflation': AI Devours Global Memory Capacity and Consumers Foot the Bill", 2026-07-22. https://news.cnyes.com/news/id/6542427 17. [CNA#1587729] Central News Agency, "iPhone Prices in Japan Up 10%, Suspected Memory Cost Increases Plus Yen Depreciation", 2026-07-18. https://www.cna.com.tw/news/aopl/202607180093.aspx 18. [PRTIMES#1642113] PR TIMES / Back Market Japan K.K. (corporate release), "Following Apple's iPhone Price Revision, Unit Sales of Refurbished iPhones on Back Market Increased", 2026-07-22. https://prtimes.jp/main/html/rd/p/000000083.000103337.html 19. [PRTIMES#1754949] PR TIMES / Taurus Inc. (corporate release, promotional), "Amazon Time Sale, Limited Quantity: Even Under Soaring Memory Prices, a Laptop with 16GB DDR5 and a 512GB NVMe SSD for ¥53,980", 2026-07-30. https://prtimes.jp/main/html/rd/p/000000058.000162473.html 20. [cnyes#1571086] cnyes, "Research: Memory Crowding Out Cuts Q2 Global Smartphone Shipments 11% Year-on-Year; Samsung Beats Apple to Reclaim the Share Lead", 2026-07-17. https://news.cnyes.com/news/id/6538011 21. [CNA#1351684] Central News Agency, "CHPT Pushes Capacity Expansion for AI Demand; Bright Lab Focuses on Technology to Capture the Trend", 2026-07-07. https://www.cna.com.tw/news/afe/202607070226.aspx 22. [TWSE#1721391] Market Observation Post System (individual filings have no verifiable fixed public URL and must be retrieved there by company name and filing date; this card attaches no link), "[Chicony Power] Announcement That the Company Resolved the Expected Board Meeting Date for Its Q2 2026 Consolidated Financial Report as 4 August 2026", 2026-07-27. 23. [ANK-2026-07-06-008] Takenouchi Rin, "Taiwan's July 2026 Monthly-Revenue Wave, Attribution-Discipline Round Two", 2026-07-06. https://ainews.idaeo.ai/en/idaeo/ANK-2026-07-06-008 24. [ANK-2026-07-08-001] Takenouchi Rin, "Memory Cycle, Chapter Five — the 'Taiwan Revenue and Capital-Action Chapter'", 2026-07-08. https://ainews.idaeo.ai/en/idaeo/ANK-2026-07-08-001 25. [ANK-2026-07-29-003] Takenouchi Rin, "Macronix Q2 2026: Record Revenue NT$19.125B, EPS NT$3.91", 2026-07-30. https://ainews.idaeo.ai/en/idaeo/ANK-2026-07-29-003 26. [ANK-2026-07-03-003] Takenouchi Rin, "Taiwan's 'Q2 Revenue Record Wave' Under a Composition Test: Three Monthly Reports on the Same Day, Three Engines", 2026-07-02. https://ainews.idaeo.ai/en/idaeo/ANK-2026-07-03-003 27. [ANK-2026-07-17-006] Takenouchi Rin, "iPhone Tops Next-Purchase Intent in Japan at 55.6%: LINE Research Survey Fielded February 2026, Relayed by Taiwan's Storm Media", 2026-07-20. https://ainews.idaeo.ai/en/idaeo/ANK-2026-07-17-006

> Anti-fabrication statement: Every figure in this card is as stated in its cited source and was matched verbatim against the original span. No figure absent from the sources was generated; no cross-source figures were merged into a single sentence to manufacture a comparison; no ratio or delta was computed; no sample or estimate was treated as a census. Price figures belong to different product categories, time points and institutions and each carries its own citation. official_count_verified = 0 (no official registry was re-queried), and no Wikidata Q identifiers are attached.

📊 引用級事實單元（F-Units）

CCI Yilan's Q2 2026 filing shows quarterly revenue of NT$994 million, up 8.33% from Q1 2026 and up 40.7% from the same quarter of 2025, net profit after tax of NT$157 million (down about 3% from Q1 2026, up 125% from the same quarter of 2025) and quarterly EPS of NT$2.01
F-001 · Confidence: high · Basis: news_aggregation cnyes #1757311 Grade C · 媒體報導（非企業審查） reported/filed 2026-07-30 (Q2 2026, filing basis)
CCI Yilan's cumulative first half of 2026 shows revenue of NT$1.911 billion (up 28.7% from the first half of 2025), operating profit of NT$350 million (up 22.3%), net profit after tax of NT$319 million (up 51.6%) and EPS of NT$4.1, with both cumulative first-half net profit after tax and EPS exceeding the full nine-month 2025 total
F-002 · Confidence: high · Basis: news_aggregation cnyes #1757311 Grade C · 媒體報導（非企業審查） reported/filed 2026-07-30 (first half of 2026, filing basis)
CCI Yilan explained that Q2 2026 operating profit was NT$163 million, up 30.9% from the same quarter of 2025, and that foreign-exchange gains turning substantially positive versus the same period of 2025 drove the 125% year-on-year rise in Q2 net profit after tax; Q2 set same-period records for both revenue and profit
F-003 · Confidence: high · Basis: news_aggregation cnyes #1757311 Grade C · 媒體報導（非企業審查） reported/filed 2026-07-30 (Q2 2026)
CCI Yilan's filing basis: among the company's four product lines this year, outdoor and medical remain the two growth engines, both with double-digit first-half revenue growth versus the same period of 2025 and together accounting for 80% of total revenue; the Q3 outlook lists demand from outdoor, medical, aerospace/marine and defence products
F-004 · Confidence: high · Basis: news_aggregation cnyes #1757311 Grade C · 媒體報導（非企業審查） reported/filed 2026-07-30 (first half of 2026 / Q3 outlook)
CCI Yilan filed that the "expected board meeting date for approving the Q2 2026 financial report" is 2026/07/30 (filing output date 2026/07/23, statement date 2026/07/22)
F-005 · Confidence: high · Basis: official_statement TWSE #1656142 Grade A · 法定開示 output 2026-07-23 (statement date 2026-07-22; expected meeting date 2026-07-30)
Nanya Technology's Q2 2026 average selling prices rose more than 60% versus Q1 with sales volume flat versus Q1, driving Q2 revenue of NT$82.549 billion, up 68.2% from Q1 2026, a gross margin of 79.5% (up 11.6 percentage points from Q1), net profit after tax of NT$50.192 billion and record EPS of NT$14.66; cumulative first-half net profit after tax was NT$76.25 billion, EPS NT$23.38 and net asset value per share NT$93.49
F-006 · Confidence: high · Basis: news_aggregation CNA #1415184 Grade U · 未分類／不可主張已驗證 investor call 2026-07-10 (Q2 2026 / first half of 2026)
Nanya Technology president Lee Pei-ing said that in the first half, revenue from products applied to AI infrastructure and servers already exceeded 20% of total revenue; the company continues to supply DDR5, LPDDR5 and DDR4, with DDR5 at roughly 10% of revenue and LPDDR5 scheduled for qualification in the second half
F-007 · Confidence: medium · Basis: news_aggregation CNA #1415184 Grade U · 未分類／不可主張已驗證 investor call 2026-07-10 (first half of 2026)
Lee Pei-ing said strong AI server and cloud data-centre demand is driving continued growth in HBM, server DDR5 and other high-capacity requirements and is also crowding out general-purpose DRAM capacity for smartphones, PCs, automotive and consumer electronics; memory tightness is no longer confined to high-end products, with DDR5, DDR4 and even DDR3 seeing supply tightness and price increases, and undersupply expected to last more than several quarters
F-008 · Confidence: medium · Basis: news_aggregation cnyes #1419943 Grade U · 未分類／不可主張已驗證 investor call 2026-07-10 (forward-looking and current-status remarks)
Lee Pei-ing said Q3 market prices remain higher than Q2 and that Q3 and Q4 operations look set to keep improving; existing capacity is close to full, sales volume is unlikely to rise materially before the new fab ramps, and subsequent growth will come mainly from higher memory prices and an improved product mix; whether the rate of increase converges depends on the supply–demand gap and end markets' cost tolerance, with some low-priced, small end products possibly reducing capacity or cutting purchases, and prices of products such as smartphones and televisions possibly adjusted as memory costs rise
F-009 · Confidence: medium · Basis: news_aggregation cnyes #1419943 Grade U · 未分類／不可主張已驗證 investor call 2026-07-10 (company outlook)
Nanya Technology expects capital expenditure of over NT$50 billion this year, with an internal preliminary plan to raise it to over NT$200 billion next year (that budget still requires board approval); the new fab's final full monthly capacity is planned at 45,000 wafers with total capital expenditure estimated at about NT$480 billion, and the amount invested so far is close to but has not yet reached NT$100 billion; phase-one monthly capacity is planned at 30,000 wafers in 2028 and 36,000 wafers in 2029; long-term supply agreements (LTAs) already cover about 50% of capacity
F-010 · Confidence: medium · Basis: news_aggregation cnyes #1419943 Grade U · 未分類／不可主張已驗證 investor call 2026-07-10 (2026/2027 plans; 2028-2029 capacity schedule)
Lee Pei-ing said NVIDIA has been a Nanya Technology customer for years and that Google, Microsoft, Intel and AMD are also customers; the long-term-contract share stands at 50% and short-term contracts will convert to long-term ones as they expire; the new fab is estimated to need 1,500 staff, half of whom have been hired
F-011 · Confidence: medium · Basis: news_aggregation CNA #1415184 Grade U · 未分類／不可主張已驗證 investor call 2026-07-10
TrendForce forecasts Q3 2026 server DRAM contract prices up 13-18% from Q2 2026 with the rate of increase converging, because some suppliers raised quotes early in Q2 and several US CSPs have signed multi-year long-term agreements limiting room to raise prices; its preliminary calculation puts aggregate supplier RDIMM bit supply growth in 2027 at only about 15-20% year on year, far behind server CPU unit growth; it expects contract prices to keep rising quarter by quarter from the second half of 2026 through the second half of 2027 while the rate of increase converges
F-012 · Confidence: medium · Basis: news_aggregation cnyes #1377472 Grade U · 未分類／不可主張已驗證 reported 2026-07-09 (Q3 2026 forecast; 2027 supply calculation)
BofA Global Research's late-July 2026 channel check: more tier-two OEMs and module makers have accepted a further 15-20% rise in July PC DRAM contract prices versus June, and Q3 DRAM average selling prices may rise 30-40% (the report explicitly noting this is clearly higher than TrendForce's earlier 13-18% forecast); 16Gb DDR5 spot at about US$50 and DDR4 spot at about US$80 (non-public market averages), 64GB server module contract prices above US$1,000 and the DDR5 version at about US$1,400; BofA's model puts the pace of DRAM average-price increases at 53% in Q2, 21% in Q3 and 7% in Q4, declining quarter by quarter
F-013 · Confidence: medium · Basis: news_aggregation cnyes #1724163 Grade C · 媒體報導（非企業審查） reported 2026-07-28 (late-July channel check; model estimates for Q2-Q4 2026)
South Korean customs data: the country's exports in the first 20 days of July 2026 were US$54.9 billion, up 52.3% from the same period of 2025, of which semiconductor exports were US$22.1 billion, up 180.6% from the same period of 2025, a 40.3% share
F-014 · Confidence: medium · Basis: news_aggregation cnyes #1724163 Grade C · 媒體報導（非企業審查） 1-20 July 2026 (South Korean customs data, reported 2026-07-28)
South Korea's Chosun Ilbo reported that spot quotes for 64GB server DRAM (RDIMM) had broken US$3,100, about 146% above the end-June contract price of US$1,380; per a Meritz Securities report, spot prices entered a steep uptrend from mid-July and validation demand from new AI data centres has tightened even sample memory supply; past transmission shows spot leading contract prices by about four to six weeks, with the traditional spot premium at 10% to 20%
F-015 · Confidence: medium · Basis: news_aggregation cnyes #1671952 Grade C · 媒體報導（非企業審查） reported 2026-07-24 (spot quotes vs end-June 2026 contract price)
TrendForce data: conventional DRAM contract prices rose 93% to 98% quarter on quarter in Q1 2026 and a further 58% to 63% in Q2, with NAND Flash contract prices up 70% to 75% over the same period; a year ago a mainstream PC memory module kit sold for about US$75, whereas the same specification now reaches as much as US$460
F-016 · Confidence: medium · Basis: news_aggregation cnyes #1641000 Grade C · 媒體報導（非企業審查） Q1/Q2 2026 contract prices; the PC module price compares the report date of 2026-07-22 with "a year ago"
Market participants say this price surge did not originate in supply-chain disruption or a shortage of fabs but in memory makers deliberately allocating resources to the AI market for profitability reasons; HBM both sells for more than conventional DRAM and consumes more wafer capacity, so each additional HBM wafer further reduces available consumer DRAM; per recent research houses and supply-chain information, HBM capacity is almost entirely sold out through the end of 2026 and most 2027 capacity is already pre-booked by major customers
F-017 · Confidence: medium · Basis: news_aggregation cnyes #1641000 Grade C · 媒體報導（非企業審查） reported 2026-07-22
ADATA chairman Chen Li-bai said in a press release that memory suppliers have notified the company that Q3 DRAM contract prices will rise a further 20% to 30% and NAND Flash by 35% to 40%; ADATA also stated that rising DRAM and NAND Flash prices were the main driver behind its record June revenue
F-018 · Confidence: medium · Basis: news_aggregation CNA #1339811 Grade C · 媒體報導（非企業審查） reported 2026-07-06 (Q3 contract-price notification; attribution for June 2026 revenue)
Transcend said that as AI and edge computing flourish, industrial automation, in-vehicle devices and high-performance computing continue to drive robust demand for high-capacity, high-reliability storage modules, and it holds a cautiously optimistic view of second-half market demand
F-019 · Confidence: medium · Basis: news_aggregation CNA #1339811 Grade C · 媒體報導（非企業審查） reported 2026-07-06 (June 2026 / second-half outlook)
Winbond attributed its record June revenue "mainly to the effect of market supply and demand"; Macronix attributed its record June revenue "mainly to the effect of increased market demand"; Team Group said steady order momentum in industrial-control and system-integration markets plus the gradual benefits of product-mix optimisation and niche-market expansion were the main drivers of June revenue growth
F-020 · Confidence: high · Basis: news_aggregation CNA #1351917 Grade C · 媒體報導（非企業審查） reported 2026-07-07 (June 2026, self-tallied)
Chey Tae-won, chairman of South Korea's SK Group and of the Korea Chamber of Commerce and Industry, said in an interview with the Korean outlet EtNews that current memory prices are "abnormal" and that "memory prices should come down," warning that persistently high prices would ultimately push up PC and smartphone prices and trigger "chip inflation"; he urged the industry to voluntarily lower margins and expand supply, and estimated AI-related demand would grow 60% to 100% next year while supply-side growth lags far behind; he also predicted that even if memory shortages ease, the AI industry's bottleneck will shift to energy infrastructure
F-021 · Confidence: medium · Basis: news_aggregation cnyes #1613951 Grade C · 媒體報導（非企業審查） reported 2026-07-20 (interview remarks; "next year" = demand estimate for 2027)
Gartner coined the term "Memflation" in a report released in April 2026 (proposed by senior principal analyst Rajeev Rajput) to describe soaring memory prices and non-AI industries being forced to bear the cost pressure; Gartner forecasts total global semiconductor industry revenue to exceed US$1.3 trillion in 2026, up 64% from 2025, with global memory revenue jumping from US$216.3 billion in 2025 to US$633.3 billion, DRAM prices up 125% and NAND flash memory up 234%; although AI chips account for only about 30% of global semiconductor revenue in 2026 they dominate almost all industry growth, while the remaining 70% of the market (smartphones, PCs, automotive, industrial and appliances) must fight over residual capacity and absorb price increases; Rajput said "Memflation may destroy, or at least delay, the recovery of non-AI demand," and Gartner explicitly says meaningful price declines will come no earlier than the second half of 2027
F-022 · Confidence: medium · Basis: news_aggregation cnyes #1643449 Grade C · 媒體報導（非企業審查） Gartner report of April 2026 (relayed in reporting of 2026-07-22; 2026/2027 forecasts)
Market participants estimate that the main memory and flash memory cost for a flagship smartphone has risen from about US$40 in 2024 to US$80-100 this year; for AI-capable PCs the DDR5 memory module cost increase reaches 70-90%; Apple raised product prices across the board for the first time in June this year, with the MacBook Pro, MacBook Air, iPad Pro and iPad Air lines up by US$100 to US$300; research house Omdia's statistics show more than 70% of channel partners reported price increases this year and more than 90% faced delivery delays
F-023 · Confidence: medium · Basis: news_aggregation cnyes #1643449 Grade C · 媒體報導（非企業審查） reported 2026-07-22 (2024 versus "this year" = 2026 estimate; Apple's repricing was June 2026)
Apple raised iPhone prices in the Japanese market on 17 July 2026: the iPhone 17 entry price was reset to ¥142,800, a 10% increase (the standard model up ¥13,000), the top-end iPhone 17 Pro Max rose ¥20,000 to ¥214,800 and the affordable iPhone 17e broke ¥100,000; Apple did not raise iPhone prices in its main US market; owing to the rapid rise in memory prices Apple had already raised Japanese prices for Macs and iPads by roughly 20% to 30% in late June; CEO Tim Cook said in a June Wall Street Journal interview that with memory prices high, "product price increases are unavoidable"
F-024 · Confidence: high · Basis: news_aggregation CNA #1587729 Grade C · 媒體報導（非企業審查） reported 2026-07-18 (repricing date 2026-07-17; Mac/iPad repricing in late June 2026)
Back Market Japan K.K. announced that Apple implemented an across-the-board revision of iPhone series (SIM-free) prices at Apple's official store in Japan on 18 July 2026; over the three days from 18 to 20 July after the increase, unit sales in the iPhone-14-and-above category on the company's Japanese site rose 41.64% versus the same period the previous week (11 to 13 July), with the top three models being the iPhone 17 up 900% week on week, the iPhone 16 Pro MAX up 450% and the iPhone 16 up 156.25%; price examples (as of 22 July 2026, grade-C goods) were iPhone 17 256GB at ¥142,800 new versus ¥117,730 and up refurbished, and iPhone 16 128GB at ¥124,800 new versus ¥93,000 and up refurbished
F-025 · Confidence: medium · Basis: official_statement PRTIMES #1642113 Grade B · 平台審查企業發稿 released 2026-07-22 (sales data 2026-07-18 to 07-20, compared with 07-11 to 07-13 the previous week)
Taurus Inc.'s promotional corporate release of 30 July 2026 uses "under soaring memory prices" as its framing and cites third-party forecasts: TrendForce forecasts general-purpose DRAM contract prices up 58-63% quarter on quarter in Q2 2026 and NAND Flash up 70-75%; Gartner indicates DRAM and SSD prices combined may rise 130% by the end of 2026, potentially pushing PC prices up 17% versus 2025
F-026 · Confidence: medium · Basis: official_statement PRTIMES #1754949 Grade B · 平台審查企業發稿 released 2026-07-30 (forecasts cited are TrendForce's for Q2 2026 and Gartner's through end-2026)
Counterpoint Research reports global smartphone shipments in Q2 2026 down 11% from the same quarter of 2025, the lowest second-quarter level since 2013, with memory supply and rising costs among the main factors affecting the market in the quarter; Samsung returned to global number one with a 24% shipment share; Apple's Q2 shipments rose 3% from the same quarter of 2025 and its shipment share climbed to 20%, a record for any second quarter, while it was also the only major OEM not to raise smartphone prices in the quarter; Xiaomi, OPPO and vivo all posted double-digit declines in Q2 shipments versus the same quarter of 2025; Counterpoint maintains its forecast of global smartphone shipments down about 14% in full-year 2026 versus 2025 and expects memory shortages to possibly persist into 2027
F-027 · Confidence: medium · Basis: news_aggregation cnyes #1571086 Grade C · 媒體報導（非企業審查） reported 2026-07-17 (Q2 2026 shipments; full-year 2026 forecast)
Largan's first half of 2026 shows revenue of NT$29.209 billion, a gross margin of 49.41% (down 4.77 percentage points from the first half of 2025), net profit after tax of NT$10.793 billion (up 44.38% from the first half of 2025) and first-half EPS of NT$82.35; Q2 2026 revenue was NT$13.665 billion with a gross margin of 49.41% in line with Q1 (down 4.22 percentage points from the same quarter of 2025), net profit after tax of NT$4.67 billion (down 35.92% from Q1 2026, up 3.52 times from the same quarter of 2025) and quarterly EPS of NT$35.72
F-028 · Confidence: high · Basis: news_aggregation cnyes #1375920 Grade U · 未分類／不可主張已驗證 investor call 2026-07-09 (first half of 2026 / Q2 2026)
Largan CEO Lin En-ping said July momentum was slightly better than June and August momentum better than July, with capacity also expected to be full; variable-aperture lenses entered mass production in Q2 but shipment volumes are small and initial yields were poor, so whether assembly yields improve in Q3 is the major swing factor for Q3 gross margin; memory and raw-material costs this year have already forced major US brands and others to raise prices of 3C electronics and may cause key customers to revise total sales volumes down, yet key customers have not revised down their estimated lens order quantities for this year's new models
F-029 · Confidence: medium · Basis: news_aggregation cnyes #1380644 Grade U · 未分類／不可主張已驗證 investor call 2026-07-09 (July/August 2026 momentum, Q3 outlook)
Largan chairman Lin En-ping said "we are doing CPO so as not to be wiped out by AI"; at a recent investor call he confirmed winning the company's first fibre array (FA) mass-production order, with samples shipped in July, trial production at the end of Q3 and mass production as early as the middle of next year, while advancing proofs of concept (POC) with multiple customers; internal alignment precision has reached below 0.3 micrometres, better than the market level of about 0.5 to 0.8 micrometres; on technology paths Corning focuses on EC (edge coupling) while Largan is positioned on GC (grating coupling), the current market mainstream; the report states explicitly that CPO remains a small share of near-term revenue
F-030 · Confidence: medium · Basis: news_aggregation CNA #1444129 Grade U · 未分類／不可主張已驗證 reported 2026-07-11 (order and sampling schedule; mass production no earlier than mid-2027)
TrendForce (Jibang) press release of 30 July 2026: 2027 DRAM supply will stay tight with prices firm and demand expected to exceed supply by about 1% to 2%; most DRAM new-fab wafer-input ramps fall in the second half of 2027, with a significant output contribution only in 2028; HBM wafer input is far higher than for general-purpose DRAM, constraining overall 2027 DRAM bit-supply growth; NAND Flash supply may loosen in the second half of 2027 with prices facing correction pressure, though a breakthrough in agentic AI adoption lifting SSD demand could bring overall supply and demand closer to balance; some CSPs are showing negative cash flow, so their 2027 capital-expenditure plans and any impact on memory procurement are worth watching
F-031 · Confidence: medium · Basis: news_aggregation CNA #1754333 Grade C · 媒體報導（非企業審查） press release 2026-07-30 (2027/2028 outlook)
Chunghwa Precision Test Tech president Huang Shui-ke said on 7 July 2026, at Taiwan's venture capital and private equity annual conference, that the AI industry's demand for semiconductors and test interfaces is large and that CHPT will continue expanding capacity over the next two to three years
F-032 · Confidence: medium · Basis: news_aggregation CNA #1351684 Grade U · 未分類／不可主張已驗證 speech 2026-07-07 (capacity expansion over the next two to three years)
On 27 July 2026 Chicony Power filed that it had resolved that the expected board meeting date for its Q2 2026 consolidated financial report would be 4 August 2026
F-033 · Confidence: high · Basis: official_statement TWSE #1721391 Grade U · 未分類／不可主張已驗證 filed 2026-07-27 (expected meeting date 2026-08-04)

❓ FAQ

What did CCI Yilan's Q2 2026 financial report show, and does it match the earlier self-tallied figures?

CCI Yilan filed its Q2 report on 30 July 2026: quarterly revenue of NT$994 million (up 8.33% from Q1 2026, up 40.7% from the same quarter of 2025), net profit after tax of NT$157 million (down about 3% from Q1 2026, up 125% from the same quarter of 2025) and quarterly EPS of NT$2.01. It does not match exactly the self-tallied basis recorded in the target card (Q2 NT$995 million, first half NT$1.912 billion) — the filing basis is NT$994 million / NT$1.911 billion. The two are different bases (a self-tally is a company-released figure, not an audited report). This card places both side by side with their own citations, does not claim either corrects the other, and does not compute the delta; the reason for the difference is not stated in the source, nor is the audit/review status of this quarter's report (cnyes#1757311; see Basis item 1).

Does the comparison the target card refused — "already close to last year's nine-month profit total" — now hold?

It does, but in the form of profit against profit: CCI Yilan's cumulative first-half net profit after tax was NT$319 million, up 51.6% from the first half of 2025, with EPS of NT$4.1, and both cumulative first-half net profit after tax and EPS exceeded the full nine-month 2025 total. What the target card refused was the monthly-report comparison that mixed a revenue basis with a profit basis. Today's filing puts the comparison back on one basis (net profit after tax against net profit after tax, EPS against EPS), which is what makes it usable (cnyes#1757311).

Was Nanya Technology's record revenue driven by AI after all?

The company stated that "in the first half, revenue from products applied to AI infrastructure and servers already exceeded 20% of total revenue" — so AI attribution moved from the target card's "no AI wording in the source" to over 20% as stated by the company. But the same call made the direct driver of the record explicit: price. Q2 average selling prices rose more than 60% versus Q1 with sales volume roughly flat versus the prior quarter, driving Q2 revenue of NT$82.549 billion, up 68.2% from Q1 2026, a gross margin of 79.5% (up 11.6 percentage points from Q1) and record EPS of NT$14.66. "Over 20%" is company-stated and not third-party verified, and it does not mean that the 621.34% year-on-year rise in June revenue came entirely from AI; the company itself said that before the new fab ramps, subsequent growth will come mainly from higher prices and an improved product mix (CNA#1415184, cnyes#1419943; see Basis item 3).

How much have memory prices actually risen, and why do the numbers differ so much between houses?

Because these are five different observation points, not one number: TrendForce forecasts Q3 2026 server DRAM contract prices up 13-18% from the prior quarter; BofA's channel check says July PC DRAM contract prices rose 15-20% versus June and Q3 average selling prices may rise 30-40% (with the report explicitly noting this exceeds TrendForce's 13-18% forecast); ADATA relays a supplier notification of a further 20-30% rise in Q3 DRAM contract prices and 35-40% for NAND; South Korean media report 64GB server DRAM spot above US$3,100, about 146% above the end-June contract price of US$1,380 (the traditional spot premium being 10-20%); and TrendForce's realised series shows conventional DRAM contract prices up 93-98% quarter on quarter in Q1 2026 and a further 58-63% in Q2. They belong to different product categories (server DRAM / PC DRAM / NAND / spot), different time points and different institutions, and must not be converted into one another, averaged, or combined into a single rate of increase (cnyes#1377472, cnyes#1724163, CNA#1339811, cnyes#1671952, cnyes#1641000; see Basis item 4).

Have Japanese end-prices already been affected by memory price increases?

They have already moved, but the attribution must be graded: Apple raised iPhone prices in the Japanese market on 17 July 2026, resetting the iPhone 17 entry price to ¥142,800, a 10% increase, with the top-end iPhone 17 Pro Max up ¥20,000 to ¥214,800 and the iPhone 17e breaking ¥100,000. Apple did not raise iPhone prices in its main US market. In addition, owing to the rapid rise in memory prices, Apple had already raised Japanese prices for Macs and iPads by roughly 20% to 30% in late June. The report's headline uses the word "suspected": the attribution for this Japanese repricing is outside inference (the Nihon Keizai Shimbun noted that observers see it as reflecting yen depreciation, with the dollar–yen rate above ¥162 and more than ¥10 weaker than a year earlier) plus Cook's general remark in a June Wall Street Journal interview that "product price increases are unavoidable" — not Apple's official item-by-item attribution. The relative weights of the two factors are not quantified in the source (CNA#1587729; see Basis item 6).

Is there data on how Japanese consumers responded?

There is a single-platform signal, but it is not a market census: Back Market Japan self-published that after Apple's across-the-board revision of iPhone (SIM-free) prices at its official Japanese store on 18 July 2026, unit sales in the iPhone-14-and-above category on that platform rose 41.64% over the three days from 18 to 20 July versus the same period the previous week (11 to 13 July), with the iPhone 17 up 900% week on week, the largest gain. This is the company's own platform sales data, self-published by the company and not third-party verified; it may be treated only as a single-platform signal of Japanese substitution behaviour and must not be scaled up to the behaviour of the Japanese market as a whole (PRTIMES#1642113; see Basis item 7).

Will end-price increases directly cause demand to collapse?

Under the available sources, that straight-line inference is blocked by two counter-examples. One: Counterpoint Research reports global smartphone shipments in Q2 2026 down 11% from the same quarter of 2025 (the lowest second quarter since 2013), yet Apple's Q2 shipments rose 3% from the same quarter of 2025, its share climbed to 20% — a record for any second quarter — and it was the only major OEM not to raise prices in the quarter. Two: Largan CEO Lin En-ping said memory and raw-material costs have forced major US brands and others to raise 3C prices and may cause key customers to revise total sales volumes down, yet key customers have not revised down their estimated lens order quantities for this year's new models. The cost shock has already landed on selling prices (entry-level and mid-range models were more visibly affected), but neither the component order book nor the shipments of the firm that did not raise prices has been broken through. Lin was speaking about "estimated lens order quantities for this year's new models," not realised shipments, and he left open the possibility of rolling downward revisions if sales disappoint (cnyes#1571086, cnyes#1380644).

Is Largan still an "absentee from AI spillover"?

The coordinates have moved, but this cannot be written as "it has become an AI beneficiary stock." What is traceable: chairman Lin En-ping said "we are doing CPO so as not to be wiped out by AI," and confirmed at an investor call the company's first fibre array (FA) mass-production order, with samples shipped in July, trial production at the end of Q3 and mass production as early as the middle of next year, with internal alignment precision already below 0.3 micrometres versus the market level of about 0.5 to 0.8 micrometres. The report also states explicitly that CPO remains a small share of near-term revenue. At the same time the target card's P-003 has materialised: Largan's first-half 2026 net profit after tax was NT$10.793 billion, up 44.38% from the first half of 2025, with EPS of NT$82.35; Q2 net profit after tax was NT$4.67 billion, down 35.92% from Q1 2026 and up 3.52 times from the same quarter of 2025; and Lin said July momentum was slightly better than June, August better than July, with capacity expected to be full (CNA#1444129, cnyes#1375920, cnyes#1380644).

Will this round of memory price increases continue indefinitely?

None of the four houses says "up without pause": TrendForce says server DRAM contract prices keep rising quarter by quarter from the second half of 2026 through the second half of 2027 while the rate of increase converges; BofA's model puts the pace of DRAM average-price increases at 53% in Q2, 21% in Q3 and 7% in Q4, declining quarter by quarter; Gartner says meaningful price declines come no earlier than the second half of 2027; and TrendForce-Jibang (2026-07-30) says 2027 DRAM supply stays tight with demand exceeding supply by about 1% to 2%, while NAND Flash supply may loosen in the second half of 2027 with prices facing correction pressure. The four differ in product category and time point and cannot be combined, but they share one thing: a fork, not a slope. The Jibang release also notes that some CSPs are showing negative cash flow, so their 2027 capital-expenditure plans and any impact on memory procurement are worth watching (cnyes#1377472, cnyes#1724163, cnyes#1643449, CNA#1754333).

How does this card relate to target card ANK-2026-07-06-008?

This is its round three, doing three things the target card could not do on the day: (1) moving CCI Yilan's self-tallied basis onto the filing basis of 2026-07-30, so that the "profit total" comparison the target card refused now holds as profit against profit; (2) filling the causal gap that card declared — Nanya Technology stating at its 2026-07-10 call that AI is crowding out general-purpose DRAM capacity, and quantifying the AI share at "over 20%"; (3) filling the Taiwan–Japan gap that card declared — Apple's 2026-07-17 increase in Japanese iPhone prices and a single-platform signal of Japanese substitution behaviour. It also materialises the target card's P-003 (the Largan and Nanya Technology investor calls) and states plainly that P-002 has not materialised (Chicony Power's expected board meeting date for its Q2 report is 2026-08-04, and this card does not prejudge the outcome). Not one of the target card's judgements is overturned here — the layers were simply added once the bases matched. ---

🧠 編輯判斷（J-Units）

The methodological gain of attribution-discipline round three is that a self-tally and a filing are two bases, and only side by side do they form complete evidence. CCI Yilan's self-tally (Q2 NT$995 million / first half NT$1.912 billion, as recorded in the target card) and its filing (NT$994 million / NT$1.911 billion, cnyes#1757311) do not match exactly, and the same company even used two engine narratives between its monthly report and its quarterly filing (monthly: defence/outdoor/medical; filing: outdoor and medical as the two engines, with defence as a Q3 outlook demand source). The value of an extension card is not in overturning but in waiting for the bases to match before letting a comparison through — the "already close to last year's nine-month profit total" line the target card refused now holds today as profit against profit (first-half net profit after tax of NT$319 million and EPS of NT$4.1 both exceeding the full nine-month 2025 total)
Confidence: medium
AI attribution boundaries move, but not as "the source didn't say it, then it did" — rather as "quantified up to a specific cell." Nanya Technology went from the target card's "no AI wording in the source" to the 2026-07-10 call's "first-half AI infrastructure and server product revenue already over 20% of total" [F-007] — the boundary moved from zero to over 20% as stated by the company. Yet the company simultaneously drew another line: the direct driver of the record is price (Q2 average selling prices up more than 60% versus Q1, sales volume flat versus the prior quarter) [F-006], and before the new fab ramps "subsequent growth will come mainly from higher memory prices and an improved product mix" [F-009]. A company with over 20% AI-related revenue still describes its own growth engine as price
Confidence: medium
The causality the target card would not extrapolate has now been voiced by the supply side itself — but it remains a company basis, not demonstrated causality. Lee Pei-ing said AI server and cloud data-centre demand is driving HBM and server DDR5 growth and "is also crowding out general-purpose DRAM capacity for smartphones, PCs, automotive and consumer electronics" [F-008]; third parties describe the same structure as "supply bifurcation" (Gartner: AI chips are only about 30% of 2026 global semiconductor revenue yet dominate nearly all growth) [F-022] and as "makers deliberately allocating capacity for profitability" (market participants) [F-017]. The three sources are of one type but belong to three levels of evidence — company, institution and unnamed relay — and this card records them by level rather than merging them into a single conclusion
Confidence: medium
"How much have memory prices risen" has no single answer, only five readings from five observation points: a third party's server DRAM contract-price forecast of 13-18% [F-012]; an investment bank's PC DRAM channel check (July contract prices up 15-20% versus June, Q3 ASP 30-40%, with the report explicitly noting it exceeds the former) [F-013]; a supply-chain notification (Q3 DRAM 20-30%, NAND 35-40%) [F-018]; a spot-to-contract premium of 146% as stated in the original report (traditionally 10-20%) [F-015]; and the realised contract-price series (Q1 2026 up 93-98% quarter on quarter, Q2 58-63%) [F-016]. They belong to different product categories, time points and institutions, and any averaging or combination would generate a figure that exists in no source — the correct reading of this chain is "five readings," not "one rate of increase"
Confidence: medium
With the Taiwan–Japan gap filled, the structure visible is that cost has already landed on end prices, but the distribution of demand destruction is extremely uneven. On the Japan side: Apple raised iPhone prices on 2026-07-17 (iPhone 17 entry price ¥142,800, a 10% increase), had already raised Japanese Mac and iPad prices by roughly 20-30% in late June, and did not raise prices in its main US market [F-024]; substitution behaviour produced a single-platform signal (Back Market Japan's own platform saw three-day unit sales up 41.64% versus the same period the previous week) [F-025]; and retail has already written "under soaring memory prices" into its sales pitch [F-026]. But the demand side simultaneously offers two counter-examples: Counterpoint records global Q2 2026 smartphone shipments down 11% from the same quarter of 2025, while Apple, which did not raise prices, saw shipments up 3% year on year and a 20% share, a record for any second quarter [F-027]; and Largan says key customers have not revised down estimated lens order quantities for this year's new models [F-029]. Between pass-through and destruction lie each firm's pricing and supply choices
Confidence: medium
That three actors each coined a term (Chicony Power's "tech inflation," Chey Tae-won's "chip inflation" [F-021] and Gartner's "Memflation" [F-022]) does not mean they cite one another, but it does mean the same phenomenon has been separately named by a downstream manufacturer, an upstream group chairman and a research house — naming density is itself a structural signal. And all four houses' timelines describe a fork, not a slope: TrendForce says the rate of increase converges [F-012], BofA's model says the pace of average-price increases declines quarter by quarter (Q2 53%, Q3 21%, Q4 7%) [F-013], Gartner says meaningful declines come no earlier than the second half of 2027 [F-022], and TrendForce-Jibang says 2027 DRAM stays tight while NAND may loosen in the second half [F-031]
Confidence: medium
Between AI absence and AI benefit there is one more cell: betting in order not to be eliminated. In the target card Largan was an absentee (June revenue at a 13-month low, no AI attribution in the source). Three days later its chairman described his motive for entering CPO as "we are doing CPO so as not to be wiped out by AI," and the company won a first FA mass-production order (samples in July, trial production at the end of Q3, mass production as early as mid-next-year), while the report states explicitly that CPO remains a small share of near-term revenue [F-030]. This cell cannot be written as "already an AI beneficiary stock," but it can no longer be written as "absent" either — it is evidence of a course change forced by AI's gravitational field
Confidence: medium

🔮 待驗證假設（P-Units）

The actual impact of Chicony Power's "tech inflation" on its own profits — the company filed on 2026-07-27 that the expected board meeting date for its Q2 consolidated financial report is 2026-08-04 [F-033]. No financial report is available for verification as of this card's filing date, and this card does not prejudge the outcome (open)
Status: open
The materialisation of CCI Yilan's Q3 outlook (customer orders and shipments on a normal cadence, with continued contributions from outdoor, medical, aerospace/marine and defence demand) and of the target card's record that capacity is near full through end-October 2026 with some long-term orders visible into Q3 2027 — to be verified against Q3 2026 monthly revenue and the Q3 report (open)
Status: open
Whether Nanya Technology's over-NT$200-billion capital expenditure for next year (2027) receives board approval, and the materialisation of the phase-one monthly capacity of 30,000 wafers in 2028 [F-010] (open)
Status: open
Largan's CPO: materialisation of the first FA mass-production order from July sampling through end-Q3 trial production to mass production as early as mid-next-year, and whether assembly yields improve in Q3 (which the company calls the major swing factor for Q3 gross margin) [F-029][F-030] (open)
Status: open
Whether the 2027 fork holds — TrendForce-Jibang's "DRAM demand exceeding supply by about 1% to 2%, NAND possibly loosening in the second half" [F-031] and Gartner's "meaningful declines no earlier than the second half of 2027" [F-022] both require verification against actual 2027 contract prices (open)
Status: open
Whether Japanese substitution behaviour expands from a single-platform signal into censusable market behaviour — Back Market Japan's self-reported three-day data [F-025] requires verification against multiple platforms or official statistics (open)
Status: open

Verification Record

AI-structured, registry-verified, editorially sampled — Kirishima Rei (AI Editorial Desk)

⚠️ This article was AI-structured (not authored by an individual human reporter).

Cross-verified by multiple AI models.

Related Analyses