Financials

Financials — record profits, no cash, peak-cycle price

Goldkey is a small Taiwanese memory-module and IC-design house whose numbers only make sense once you accept what it is: a thin-margin, commodity-priced participant in the DRAM/NAND cycle, freshly listed on the TWSE in August 2025 and now riding the sharpest memory up-cycle in a decade. In the last four reported quarters it printed the best profits in its history — full-year FY2025 net income of NT$446M and a Q1 FY2026 that alone earned more than any prior full year [1] [2]. And in the same window it burned cash on an industrial scale — FY2025 operating cash flow of negative NT$1,774M against that NT$446M of accounting profit [3].

That contradiction is the whole story. The core question this page answers: is the FY2025–Q1 FY2026 earnings explosion a durable step-change, or a cyclical peak funded by debt and inventory that will reverse when memory prices roll over?

Sources: FY2025 income statement, cash-flow statement and balance sheet [4] [5]; ROE per FY2025 Annual Report chairman's letter [6].

What kind of business this is — a price-taker, not a compounder

Goldkey designs and assembles DRAM ICs and JEDEC memory modules, flash ICs, SSDs, USB drives and memory cards, selling under its own Neo Forza brand and through ODM/SI services. In FY2025, DRAM products were 81% of revenue (NT$6,234M) and flash products 19% (NT$1,462M) [7]. It buys memory die, adds modest assembly and design value, and resells — so its gross margin is set by the spread between memory spot prices and its purchase cost, not by pricing power. That single fact explains everything volatile in the statements below: when memory prices rise faster than inventory cost, margins spike; when they fall, margins collapse toward zero.

For a beginner: a "memory-module maker" like this sits close to a distributor in economics. Contrast that with a chip designer that owns proprietary IP — Goldkey owns some IC design, but the bulk of its cost is bought memory, so it behaves like a commodity converter. Judge it on through-cycle normalized earnings and cash conversion, never on a single peak year.

The seven-year record: a violent cycle now at its top

Here is the standard multi-year statements view. Read the margin columns first — they are the tell.

All NT$ millions except EPS. Sources: FY2019–FY2023 from the FY2023 Annual Report five-year condensed income statement [8]; FY2024–FY2025 income from the FY2025 report [9]; cash flows FY2024–FY2025 [10]; earlier-year cash flows derived from reported filings.

Three things jump out. First, revenue is not a growth series — it is a cycle: NT$5.05B in FY2019, down to NT$3.58B by FY2022, back to NT$7.70B in FY2025 [11]. FY2019's revenue was higher than any year until FY2024. Second, gross margin oscillates between roughly 2% and 10% and even turned the operating line negative in FY2022 (operating loss of NT$27M) [12]. Third, FY2025's 10.0% gross margin is the best in the entire record [13]. Management itself frames FY2025 as the start of "a new growth cycle" for the memory industry driven by cloud-to-edge and AI demand [14] — but the seven-year shape argues that "cycle" is the operative word.

Source: gross, operating and net margins derived from reported income statements, FY2023 Annual Report five-year table [15] and the FY2025 report [16].

The Q1 FY2026 blow-off

The most recent quarter is where the cycle goes vertical. Q1 FY2026 (Jan–Mar 2026) revenue was NT$3,958M, up 178% year-on-year, gross margin exploded to 30.4% (from 5.0% a year earlier), operating margin hit 27.4%, and net income was NT$856M — more than the entire FY2025, at basic EPS of NT$10.76 for a single quarter [17]. A 30% gross margin in a business that averaged mid-single-digits for six years is not a new normal; it is what a supply-squeezed memory spike looks like at the module level. The right way to hold this number is as evidence of cycle amplitude, not of structural quality.

Earnings quality: the profit-to-cash chasm

This is the crux of the entire investment case, so it gets the most room. A memory converter grows by buying inventory ahead of demand and extending receivables — so in an up-cycle, reported profit and operating cash flow move in opposite directions. Goldkey is the textbook case.

Source: net income and cash-flow statements, FY2024–FY2025 report [18] [19]; earlier years derived from reported filings.

In FY2025 the reconciliation is stark: pre-tax profit of NT$561M was overwhelmed by a NT$1,879M increase in inventory and a NT$600M increase in receivables, driving cash used in operations to NT$1,774M [20]. Because capex was minimal (NT$27M), free cash flow was roughly NT$-1,801M. Inventory alone quadrupled to NT$2,619M and is now half of total assets [21].

Widen the lens and the pattern is not a one-year quirk. Add up FY2019–FY2025:

Q1 FY2026 extends it. Despite NT$856M of profit, operations still used NT$245M of cash, as inventory rose another NT$1,314M and prepayments to suppliers ballooned to NT$2,748M (from NT$188M) — the company pre-paying to lock down scarce memory supply [22] [23]. There is a genuine bull tell buried in the same statement, though: contract liabilities — customer cash paid in advance — jumped to NT$1,939M from NT$66M [24]. Customers pre-paying is the strongest evidence of a tight market and a real order book — but it also means much of the working-capital build is customer-funded, which cuts both ways if orders are cancelled in a downturn.

The judgment: earnings quality is low through the cycle and negative right now. The profits are real accounting profits, but they are locked inside a fast-growing inventory position whose value depends entirely on memory prices staying high. This is the single most important thing to underwrite.

Balance sheet: from debt-free to founder-guaranteed leverage

For most of its history Goldkey ran a conservative balance sheet — it carried no long-term debt through FY2023 and equity was 60–80% of assets in FY2021–FY2022. The leverage is a recent, deliberate choice to finance the cycle.

NT$ millions. Sources: FY2025/FY2024 balance sheet [25], Q1 FY2026 balance sheet [26], and the FY2023 Annual Report five-year data [27].

The balance sheet doubled in FY2025 and grew another 64% in Q1 FY2026 to NT$8,508M [28]. By FY2025, interest-bearing debt reached roughly NT$2.6B — short-term borrowings of NT$1,221M, a new NT$938M convertible bond, and NT$402M of long-term bank loans — pushing the equity ratio down to 41% (debt-to-equity about 1.4) [29]. Net debt is roughly NT$2.4B against cash and near-cash of only about NT$255M — this is a company that runs with very little cash on hand and leans on revolving trade finance.

Two structural risks sit inside that debt. First, a chunk of it is purchase-financing and receivables-financing carrying interest rates up to 5.4% [30] — expensive, short-dated money that must be rolled continuously, exactly the funding that dries up if lenders lose confidence mid-downturn. Second, the borrowings are personally guaranteed by chair and president Tseng Chen, whose guarantee covered NT$1,344M of the company's facilities [31]. Founder guarantees are common for small Taiwanese caps, but they signal that the company's credit still rests on the controlling family rather than on standalone balance-sheet strength — and the roles of chair and president are combined in one person.

The flip side: with inventory at NT$2.6B carrying only a ~NT$117M obsolescence reserve [32], the balance sheet is a leveraged bet on memory prices. If prices hold, that inventory converts to high-margin sales and cash floods in; if they fall, the company faces both write-downs and a working-capital unwind while servicing short-term debt. The balance sheet is a source of risk, not flexibility.

Returns and capital allocation

Return metrics look excellent — but they are peak-cycle metrics. FY2025 ROE was 26.9% and ROA 12.4%, roughly double FY2024 [33]. Set that against the same company's ROE of 1.7% in FY2022 and 6.9% in FY2023 [34]. A business whose ROE swings from 2% to 27% over three years does not have a stable return on capital — it has operating leverage on a commodity spread.

On capital allocation, FY2025 was a year of raising capital, not returning it: the company completed its IPO cash raise (about NT$513M), issued a NT$995M convertible bond, and drew short-term debt — while paying a modest NT$102M cash dividend [35]. Shareholders should expect dilution, not buybacks: the convertible bond is already converting (weighted diluted shares rose to about 87M in Q1 FY2026, and FY2025 diluted EPS of NT$6.12 sits below basic NT$6.33) [36]. The cash generated in good years is not returned; it is reinvested straight back into inventory. For an investor, this means the value case rests entirely on the durability of the spread, because there is no capital-return cushion.

Valuation: cheap on peak earnings, expensive on mid-cycle

The stock has re-rated violently in step with the cycle. Since the pre-listing period it has moved from about NT$36 to a peak of NT$266 in May 2026 — roughly a 7x move — before pulling back about 37% to NT$167 by 1 July 2026.

Source: TWSE daily closing prices (month-end), company market data as reported.

Here is the memory-cycle valuation trap in numbers. On the last twelve months of earnings (roughly NT$16.6 EPS after the Q1 FY2026 spike), the stock trades at only about 10x trailing P/E — superficially cheap. But that denominator is peak earnings. On FY2025's full-year EPS of NT$6.33 the multiple is ~26x, and on book value the stock trades at roughly 4.0x (book value per share about NT$42) — a rich multiple for a commodity converter [37].

Source: price per TWSE market data; multiples derived from reported EPS and equity [38] [39].

The classic memory pattern is that the P/E looks lowest exactly at the earnings peak — because the market correctly refuses to capitalise peak margins. To sanity-check with normalized earnings: apply the company's own through-cycle net margin of ~3% to a normalized revenue of, say, NT$12–14B, and normalized net income is roughly NT$400–500M — putting the stock at a "mid-cycle" P/E in the mid-20s to 30s, not 10x. Nothing here is cheap once you strip out the peak. The ~37% pullback from the May high suggests the market is already beginning to price the cycle turn even as reported earnings still rise.

Peers: same cycle, thinnest margin

Goldkey's indexed peers are all genuine Taiwanese memory-module makers — a good comparison set. Every one of them is posting record margins in the same window, which both validates the cycle and frames Goldkey's relative position.

Sources: ADATA Q1 2026 investor presentation (gross margin 55.7%, EPS NT$30, net income NT$9.97B) [40]; Team Group Q1 2026 presentation (revenue NT$9,045M, EPS NT$27.00) [41]; Transcend Q4/FY2025 presentation (FY2025 gross margin 46.8%, EPS NT$12.98; Q4 gross margin 61.6%) [42] [43]; Unifosa 2025 H1 presentation (net loss) [44]. Goldkey per its Q1 FY2026 report [45].

The relative read is clear: Goldkey has the lowest gross margin of the profitable branded players — 30.4% versus 55.7% at ADATA and 61.6% at Transcend in the latest quarter [46] [47]. Peers like Transcend and ADATA have shifted more of their mix into higher-value industrial, embedded and enterprise memory with proprietary content; Goldkey remains more DRAM-module-weighted and distribution-like. That thinner margin is precisely why Goldkey has less cushion when the spread compresses — its operating line went negative in FY2022 while more differentiated peers stayed profitable. That every peer's quarterly EPS is now exceeding its prior full year is itself the loudest signal that this is a sector-wide peak, not a Goldkey-specific breakout.

What the numbers confirm, and what to watch

What the financials confirm: Goldkey is a real, growing participant in the AI-driven memory up-cycle, with genuine — if commodity — profits and an order book strong enough that customers now pre-pay. What they contradict: any notion that FY2025–Q1 FY2026 represents durable, high-quality earning power. Margins are at a seven-year peak, cash conversion is deeply negative, leverage has climbed from zero to ~NT$2.6B on founder guarantees, and returns on capital are cyclical, not structural. The valuation is a trap in both directions — cheap on peak EPS, expensive on any normalized basis.

The one metric that resolves the debate is not the income statement. It is whether the enormous inventory converts to cash without a write-down — i.e., whether gross margin and operating cash flow hold together as memory prices normalize.

The first financial metric to watch is gross margin, read alongside the inventory balance and operating cash flow. Gross margin is the direct readout of the memory-price spread that drives every line of this business; a rollback toward its 2–5% historical band — especially if it coincides with the NT$2.6B–NT$3.9B inventory position converting into write-downs rather than cash — would turn today's "record profits" into tomorrow's losses far faster than revenue growth could offset. If, instead, gross margin holds in double digits and operating cash flow finally turns positive as inventory sells through, the earnings become real and the stock is genuinely inexpensive.