Deck

Goldkey Technology · 3135 · TWSE

Goldkey Technology is a small, fabless Taiwanese memory-module maker — DRAM and flash modules, SSDs and USB drives under its Neo Forza brand and via ODM — that buys memory chips, adds assembly, and resells them.

NT$167
Share price 1 Jul 2026
NT$12.9B
Market cap
NT$7.7B
FY2025 revenue
77
Employees fabless
Priced at NT$28 in its August 2025 IPO and first traded near NT$36; the AI-memory squeeze drove a ~7x run to a NT$266 peak on 29 May 2026, then a ~37% pullback to NT$167 by 1 July 2026.
2 · The decisive fact

Record profits that have never turned into cash

  • The best quarter ever still burned cash. In Q1 FY2026 Goldkey earned NT$856M — more than all of FY2025 — and collected a NT$1,873M customer prepayment, yet operating activities still used NT$245M of cash as inventory and supplier prepayments swelled.
  • Seven years, the same pattern. Across FY2019–FY2025 the company booked about +NT$1.13B of cumulative net income but consumed roughly −NT$1.9B of operating cash; FY2025 alone turned NT$446M of profit into −NT$1,774M of operating cash flow.
  • Cash comes only when it shrinks. The one strongly cash-generative year, FY2022, was a down-cycle year when the business contracted and released working capital — profit lands as cash when Goldkey stops growing, not when it grows.
The profits are real accounting profits, locked inside a fast-growing inventory whose value depends entirely on memory prices staying high.
3 · The money picture

A genuine earnings explosion, priced at the top of the cycle

30.4%
Q1 FY2026 gross margin vs 5.0% a year earlier
NT$856M
Q1 FY2026 net income > all of FY2025
26.9%
FY2025 ROE 1.7% in FY2022
~4.0x
Price / book ~26x FY2025 EPS

Goldkey is a commodity converter: it buys memory die from a three-firm oligopoly, adds modest assembly and resells, so gross margin is set by the spread between memory spot prices and its purchase cost. That is why margins swing from ~2% to 30% across a cycle and revenue oscillates rather than compounds. At NT$167 the stock looks cheap at ~10x trailing peak EPS — but that is the memory trap: the multiple is lowest exactly at the earnings peak, and on FY2025 earnings or on book value it is expensive.

4 · The funding model

Growth paid for with debt and serial dilution

  • Debt-free to leveraged in two years. Goldkey carried no long-term debt through FY2023; by FY2025 it ran roughly NT$2.6B of interest-bearing debt — short-term trade finance at rates up to 5.4% — dropping the equity ratio to 41%.
  • Dilution is the model, not a tail risk. A first NT$1.0B convertible bond has largely converted into ~17.4M shares, about 22% of the post-IPO base; a second NT$1.5B convertible at a NT$129.9 strike was placed in May 2026.
  • Management has pre-announced the next raise. Its own FY2026 plan projects a NT$1,177M cash shortfall to be plugged by a new convertible bond, with a further NT$6–10B raise flagged — on a bank line personally guaranteed by the chair, who is also president.
The company's credit still rests on one controlling family's balance sheet as much as its own.
5 · The competitive position

The smallest, thinnest-margin name in its cluster

  • Sub-scale and not gaining. Goldkey holds about 6% of Taiwan's listed memory-module cluster against ADATA's 41%; its share has sat at 5–6% for three straight years while larger rivals move up-market.
  • No differentiation to defend the margin. R&D was just 0.39% of sales in FY2025; even at the peak its 30.4% gross margin trailed ADATA's 55.7% and Transcend's 61.6%, both of which carry far more industrial and enterprise mix.
  • A price-taker by construction. Roughly 90 cents of every revenue dollar is a memory chip whose price Goldkey does not set; its operating line went negative in the FY2022 downturn while more differentiated peers stayed profitable.
Every listed peer's quarterly EPS now exceeds its prior full year — the loudest sign this is a sector-wide peak, not a Goldkey breakout.
6 · What to watch

A real super-cycle, riding on a fragile balance sheet

  • What supports the story. Demand is customer-funded, not speculative: Q1 FY2026 contract liabilities jumped ~29-fold to NT$1,940M as customers pre-paid to lock supply, FY2025 ROE reached 26.9%, and the AI build-out is diverting memory capacity to HBM and tightening the whole market.
  • What cuts against it. The model has never self-funded a growth cycle; Goldkey is the thinnest-margin, smallest player at a seven-year margin peak, carrying rising leverage and a self-disclosed funding gap that points to more dilution.
  • What decides it. Cash conversion. Two-plus consecutive quarters of firmly positive operating cash flow with inventory normalizing would validate the bull; gross margin reverting toward its 2–5% band as inventory turns into write-downs would validate the bear.

Watchlist to re-rate: Operating cash flow turning firmly positive for two-plus straight quarters; whether gross margin holds in double digits or reverts toward 2–5%; and the size and terms of the flagged FY2026 capital raise when it lands.