Deck

Goldkey Technology Corporation · 3135 · TWSE

A small Taiwanese memory-module maker that buys DRAM and NAND flash chips from the big manufacturers, assembles and distributes them under its own Neo Forza brand, and adds an ODM arm.

$5.61
Price
~$433M
Market cap
$259M
FY2025 revenue
10.0%
FY2025 gross margin
Figures converted from Taiwan dollars at historical FX rates — see data/company.json.fx_rates. Listed at $0.94 in August 2025, the shares ran to $8.75 by late May 2026 and eased to $5.61 by July — close to six times the offer price and about 36% below the peak.
2 · The margin cycle

The record margin is a memory-price windfall the whole industry shared.

30.4%
Q1 FY2026 gross margin triple any full year
10.0%
FY2025 margin a seven-year high
2.0%
FY2022 trough margin
0.39%
R&D as % of sales

Every Taiwan memory-module maker printed a record gross margin the same quarter as DRAM contract prices roughly quintupled — ADATA at 55.7%, Team Group at 38.0%, Transcend above 60% — and Goldkey's 30.4% was the lowest of the group. The margin is an inventory holding gain on chips bought cheaply and sold into a price spike, not pricing power: the same mechanism ran in reverse to a 2.0% margin when prices fell in 2022, and Goldkey's own margin has only ever cycled between roughly 2% and 10%.

3 · Cash conversion

The record profit consumed cash rather than produced it.

$15M
FY2025 net income
-$60M
FY2025 operating cash flow
$88M
Inventory 50% of total assets
-$64M
Cumulative operating cash FY2019–FY2025

The gap is almost entirely a working-capital build — inventory more than tripled and receivables rose $20M as the company stocked chips into a rising market. Across seven years, cumulative net income of about $38M sits against roughly $64M of operating cash outflow; the model releases cash only when growth stops. The FY2025 dividend and the projected FY2026 funding gap are met with borrowing and convertible bonds, not with cash the business generated.

4 · Competitive standing

The differentiation Goldkey markets does not yet show in the numbers.

  • Smallest in the group. Roughly 6% of the listed Taiwan module market against ADATA's 41% — second-from-last of seven peers by revenue.
  • Lowest margin, even at the peak. Its 10.0% FY2025 gross margin is under half ADATA's 27.8% and under a quarter of Transcend's 46.8%, below the industry's own 20.7–31.1% range.
  • Thin research, commodity mix. R&D is 0.39% of sales across 77 employees, and in the record year the sales mix shifted further toward commodity DRAM at 80.9%, not away from it. Transcend, a 47%-margin industrial-memory specialist, is the proof of what a real moat in this business looks like.
The industrial and AI-edge direction is real and the certifications exist; a moat is not yet evident in margin, share, or returns.
5 · Valuation

The price capitalizes peak-cycle profit as if it were durable earning power.

  • Richer than a proven peer. About 27x trailing earnings and 7x book — above Transcend's ~19x and 4.7x, despite Goldkey being the smallest and lowest-margin of the group.
  • The denominator is a peak. Annualized, the Q1 run-rate is roughly 4x forward earnings — the bull case; on the mid-cycle margin its own record averages, the same shares cost 60–100x.
  • A growing share count. Two zero-coupon convertible bonds struck below market are lifting the count from 77.5m toward about 90m, so rising earnings divide across a wider base.
With DRAM contract prices still rising into late 2026, near-term earnings could be very large — a low forward multiple at a cyclical top is also the shape of a value trap.
6 · Control and concentration

The balance sheet runs on the chair's signature, and the payout is financed.

  • Personally guaranteed debt. Founder-chair Tseng Chen, who is also president, personally guarantees $63M of bank facilities; the drawn $56M equals essentially all of Goldkey's bank debt.
  • A financed dividend. The $11M FY2025 cash dividend was declared in a year that burned $60M of operating cash, with a second convertible bond named as the funding source.
  • A narrow revenue base. Two unnamed customers were 41% of FY2025 sales, and the auditor's sole key audit matter was the genuineness of that fast-growing revenue — not the inventory at half the balance sheet.
A corporate director, Tait Technology, trades on both sides of Goldkey's book; the sums are small, but purchases quadrupled year-over-year and are worth tracking.
7 · What to watch

On today's evidence the cyclical read carries more weight, and the resolvers are dated and public.

The cyclical read: a sub-scale price-taker with no established moat, a gross margin that has only ever cycled, a record year that produced no operating cash, and a distribution funded by dilution.

The other side: genuine and current — DRAM contract prices were still rising into late 2026, the customer base did widen in FY2025, and the chair's guarantee is real capital at risk.

What settles it: observable filings, not argument — whether the 30% margin holds a floor above the historical high-single digits once prices fall, and whether a quantified non-consumer revenue share ever appears.

This report is a guided study, built chapter by chapter for Goldkey — the tabs are a deliberate sequence, not a fixed template.

Watchlist to re-rate: The first monthly revenue print that turns down year-over-year after prices peak; the return of an inventory write-down charge in cost of sales; and a first disclosed non-consumer / industrial-AI revenue share.