History

History — a 27-year survivor, ten months public, riding the biggest cycle of its life

Goldkey Technology (凌航科技, TWSE 3135) is not a young company with a short story — it is a 1998-vintage memory-module house that only became a public company on 6 August 2025 [1]. So the "history" that matters here is two stories layered on top of each other: a quarter-century of surviving brutal memory cycles as a tiny, low-margin operator, and a ten-month-old public narrative that pivoted hard toward "AI memory" just as an AI-driven DRAM up-cycle lifted the numbers to record highs. What management said was going to change did change — the strategy pivot to AI/industrial-control was written down in the FY2023 letter before it paid off — and disclosures read as honest, cycle-down years included. But the banner results ride a price surge management itself credits, record FY2025 profit came with record negative operating cash flow, and there is essentially no public track record yet against which to judge them. Credibility here is unproven-but-honest, not earned.

The shape of the story

Because there are no earnings-call transcripts for Goldkey (the company releases investor decks and webcasts only, and its decks explicitly decline to forecast financials [2]), the primary record is the IPO prospectuses, five annual reports (FY2021–FY2025), and the quarterly financials. That is the corpus this page is built and cited from.

1. Who was here first: the business current leadership inherited

This team did not build Goldkey — they inherited a going concern with 20-plus years of history. The company was formally established on 30 July 1998 [3], grew through cash increases and capitalized earnings (founding capital was just NT$21.7M of an eventual NT$683.7M base) [4], and was already a ~NT$5bn-revenue, profitable business in 2019 — years before the current chapter.

Leadership anchor. The company is controlled and run by Chair & President Tseng Chen (曾珍), representing the founding-family Sheng-Yun Investment (盛耘投資) block; her board seat traces to 2011 (初次選任 100.6.14) and she has signed the shareholder letters across the entire public record [5]. The chair and CEO roles are combined, which the company defends on grounds of a lean organization and her 20-plus years of memory-industry experience [6].

The chapters. Goldkey first touched the public markets in 2021, when it registered for emerging-board (興櫃) trading under code 3135 [7]. The present strategic chapter began in 2024, when management stood up an AI/industrial-control business division and the earnings inflection started; the capital-markets chapter closed in 2025, when the TWSE approved a main-board listing on 20 June 2025 [8] and Goldkey listed on 6 August 2025, appearing as a "listed company" in its FY2025 capital table [9].

2. What "history" actually looks like here: a whipsaw, not a trend

The single most important thing a reader must internalize is that Goldkey's income statement does not trend — it whipsaws with the memory cycle. Revenue was already NT$5.05bn in 2019, fell for three years, and only broke out in FY2024–FY2025. Net income swung from NT$197M (FY2019) to NT$40M (FY2020) to NT$212M (FY2021) to NT$18M (FY2022) and back up — a 12x peak-to-trough range inside four years.

Source: FY2019–FY2023 from the FY2023 Annual Report five-year condensed income statement [10] [11]; FY2024 and FY2025 from the audited financial reports [12] [13].

Management has been consistent — and refreshingly honest — about why it swings. The FY2022 letter attributed the profit collapse (EPS NT$0.30) plainly to war, inflation and falling memory prices, while noting the company "still stayed profitable through active risk management" [14]. The FY2023 letter called the bottom explicitly: three end-markets weak, memory prices "persistently soft," and IC-maker utilization dropping below 80%, with recovery only beginning in Q4 on AI demand [15]. That is a management that names its misses rather than dressing them up — a genuine credibility positive.

The tell the P&L hides: profit and cash have gone opposite ways

Look at the same years through operating cash flow and the "record year" story cracks. FY2025 posted record net income of NT$446M — and record negative operating cash flow of NT$1,774M. The pattern is chronic: FY2022's modest profit threw off +NT$636M of operating cash, while FY2023's smaller profit burned NT$909M.

Source: net income per filings above; operating cash flow FY2019–FY2024 as reported (MOPS/audited statements) and FY2025 from the audited cash-flow statement [16].

In FY2025 the gap was plugged almost exactly by financing: NT$1,839M of net financing inflow — a convertible-bond issue, short-term borrowings, and the IPO cash increase — offsetting the operating drain [17]. Working capital, not operations, is where the profit went — the subject of Section 5.

3. The narrative drift: from "memory-module maker" to "AI memory"

Read the five annual reports in sequence and the story visibly bends. The macro-shock and pandemic language of FY2021–FY2022 fades, and AI, the AI/industrial-control division, and a DDR5 roadmap rise to the top — the clearest narrative pivot in the record.

Source: annual-report shareholder letters and risk sections, FY2021–FY2025 [18] [19] [20] [21] [22].

The arc, year by year:

• FY2021 — steady niche operator. The story is consistency through COVID: revenue ~NT$4.38bn, EPS NT$5.43, framed as staying the course [23]. Memory-price risk is present but treated as an old, familiar hazard — the report even recalls the industry's "prices change three times a day" (一日三市) folklore [24]. AI does not appear.
• FY2023 — the pivot is written down. With the business in a trough, the forward plan is re-cast entirely around AI: deepening ties to "AIoT edge computing and AI-related applications" [25]. This is the important date: the AI story was articulated before the payoff, not retrofitted onto it.
• FY2024 — the vehicle appears. Management stands up an "AI Industrial-Control Business Division" (AI 工控事業部) to supply custom modules for AI training, edge computing and industrial control [26] [27].
• FY2025 — the claim of success. The FY2025 letter says the new division "has already shown results," and frames the year ahead as a "new growth phase combining scale expansion and technology deepening" targeting AI-edge, HPC and smart industrial-control [28] [29].

The investor decks carry the same drift into the product line: the pre-listing deck framed DDR5 clock-driver modules (CKD/CUDIMM) as the answer to the AI "Memory Wall" [30], and by the 2026 decks the roadmap is explicitly an AI-computing portfolio — SOCAMM2, CAMM2/LPCAMM2, CQDIMM — pointed at "AI, networking, edge, robotics, automotive, agentic AI and medical" markets [31] [32].

The most revealing drift: the risk that flipped from threat to tailwind

Watch what happened to memory-price cyclicality. In the prospectus and every early report it is the danger — "falling average selling prices compress module-makers' margins" [33]. By the FY2025 report the same variable is written up as a benefit: rising DRAM prices from AI demand "offset part of the inflationary cost increase" [34]. The cycle didn't get less dangerous — it simply turned in Goldkey's favor. Meanwhile a new risk surfaced for the first time: rising prices force module makers to "stockpile heavily," tying up working capital in inventory [35] — the exact dynamic that produced FY2025's cash burn.

4. Promise vs delivery — and the guidance they never gave

Judging this management is unusual because they issue no financial guidance: every deck states it "does not involve any estimate or assumption of financial or operating figures" [36]. There is no quantified target to score against. What can be scored is a short list of stated commitments — and, notably, the IPO promise itself was modest: proceeds of NT$273M were earmarked entirely to replenish working capital, not a growth moonshot [37].

Source: use-of-proceeds and listing per the IPO prospectus [38] [39]; AI-strategy commitments per FY2023–FY2025 letters [40] [41] [42]; delivery per audited results [43].

On the numbers, the first public year was a genuine breakout — and Q1 FY2026 went vertical:

Source: FY2024 and FY2025 audited income statements [44] [45]; Q1 FY2026 financial report [46].

FY2025 revenue rose ~40% to NT$7,704M with net income up ~213% to NT$446M and EPS of NT$6.33 [47] [48]. Then Q1 FY2026 alone did revenue of NT$3,958M (up ~178% year-on-year) and net income of NT$856M — nearly double the entire prior fiscal year's profit in a single quarter [49]. Impressive — and exactly the vertical, top-of-cycle print that should make a memory investor cautious, not comfortable.

The uncomfortable truth about the margins

The "shift to higher value-add" is real in direction but is overwhelmingly cycle-driven. Gross margin was already ~9% in the FY2021 up-year, collapsed to 4.5% in the FY2024 trough tail, rebounded to 10% in FY2025 and leapt to ~30% in Q1 FY2026 — tracking DRAM prices far more than any structural mix change.

Source: FY2021 letter [50]; FY2024/FY2025 income statements [51] [52]; Q1 FY2026 report [53]. Values derived from reported gross profit / revenue.

5. The tell: record profit, vanishing cash, a levered balance sheet

Here is the crux of the credibility question. FY2025's headline profit of NT$446M sat on top of negative NT$1,774M of operating cash flow [54], and the shortfall was funded by NT$1,839M of new financing — convertible bonds, borrowings and the IPO cash raise [55]. Year-end cash was just NT$61M — thin for a company that quadrupled its balance sheet — and the FY2025 headline returns (ROE ~27%) are flattered by the same cyclical profit spike [56].

To their credit, management flags the mechanism itself (the stockpiling/working-capital risk, new in FY2025 [57]) and says the raise let it lower its cost of capital and optimize its debt structure [58]. That candor is why this is a warning, not a red flag.

6. Credibility verdict

Source: derived from the cited promise/delivery record in Sections 2–5.

Score: 5 / 10 — unproven, but honest. The record contains real credibility positives: management named the AI pivot in FY2023 before it paid off, described cycle-down years (FY2022, FY2023) without spin, used the IPO proceeds exactly as promised, delivered the multi-year goal of a main-board listing, and even discloses the awkward working-capital dynamic behind its own cash burn. Deloitte audits the accounts. What holds the score at the midpoint is not dishonesty but the absence of a public track record: Goldkey has been listed for roughly ten months, gives no financial guidance to be held to, and its entire breakout coincides with — and is credited by management to — an industry-wide DRAM price surge. Layer on a thin cash balance, a debt-and-equity-funded balance-sheet doubling, a combined chair/president under family control, and the memory of how fast FY2021's up-year became FY2022's collapse, and the honest verdict is: believable people, unproven as public-company stewards, in a business the cycle still controls.

7. What the story is now

The narrative today is simpler and more confident than it was in the FY2022–FY2023 trough — but not more durable. Goldkey has successfully repackaged a 27-year cyclical module business as an "AI memory" play, listed into strength, and printed the best numbers of its life. Credibility is, on balance, improving: the disclosures are consistent and honest, the strategic pivot was pre-announced and then executed, and nothing in the record reads as spin.

What to believe: the AI/industrial-control pivot is genuine and pre-dates the payoff; management tells the truth about the cycle; the listing and use-of-proceeds promises were kept. What to discount: the magnitude of FY2025–Q1 FY2026 profit as a run-rate — it is a cyclical peak on rising DRAM prices, not proof of a re-rated business; the ~27% ROE and 30% gross margin as sustainable; and any implication that the balance sheet is conservatively funded. The one thing this history cannot yet tell you — because the company has never been through it as a public company — is how management behaves when the cycle turns down. Until that chapter is written, treat the story as promising and honestly told, but unproven where it matters most.

All figures in New Taiwan Dollars (NT$) unless stated. Income-statement and balance-sheet figures are in NT$ thousands in the source filings; charts here are shown in NT$ millions.