Business
Know the Business — Auras Technology Co., Ltd. (3324.TWO)
Auras is a single-product Taiwan thermal-module pure-play whose value entirely depends on whether AI server liquid-cooling stays in the bill of materials. Revenue is up 3.0× and gross margin has more than doubled since FY2018 — but free cash flow turned NT$‑2.87B in FY2025 as the company spent the entire P&L profit, plus another billion, on capacity for the next NVIDIA platform. The market is paying ~38× trailing earnings for a business whose downside case is FY2018 commodity economics and whose upside case requires a second consecutive platform win against a much larger Taiwan peer (AVC) ramping NT$15B of capex.
The bottom line. This is not a quality compounder; it is a project-priced contract manufacturer riding a TDP-driven product cycle. The right question is not "what is the right multiple" — it is "how many platform generations can Auras stay inside NVIDIA's reference design at >25% gross margin." Get that right and the cash-flow drawdown is irrelevant. Get it wrong and the FY2018 base case (NT$7.7B revenue, 12.6% gross margin) is the gravity well the business reverts toward.
1. How This Business Actually Works
Auras sells a custom-designed thermal slot inside someone else's server, notebook, GPU or EV. There is one segment — cooling modules — and it splits into two technologies: legacy air cooling (heat pipes, vapor chambers, heat sinks, fans, ~61% of FY2024 revenue) and liquid cooling (cold plates, manifolds, CDUs, RPUs, quick disconnectors, ~39% of FY2024 revenue from AI server). Revenue is recognised per unit shipped against a specific server, PC, GPU or auto platform. There is no installed base, no recurring service, no aftermarket — every NT$ has to be re-won at the next platform refresh.
Inside one NT$100 of revenue, roughly NT$73 is raw materials and direct labour (copper, aluminium, fans, motors, TIM, water pumps, assembly in China and Thailand). About NT$6 is R&D — and that ratio has roughly doubled since FY2020 (3.30% → 6.06%) because staying inside NVIDIA's reference vendor matrix is a continuous engineering arms race. About NT$7 is SG&A, leaving NT$14 of operating profit at FY2025 mix. The point worth fixing in your head: gross margin is dragged up by liquid-cooling mix and dragged down by competitive price cuts on PC/VGA. The FY24 risk-factors line that "pressure on component manufacturers to be required to cut prices is increasing day by day" is the truthful default; the gross-margin expansion to 27.4% is the AI-cycle override.
The lift from a 12.6% gross margin in FY2018 to 27.4% in FY2025 is the entire investment thesis stamped onto one chart. Strip out the AI-server mix shift and the business reverts toward the left edge of this line — a low-teens gross-margin commoditised thermal vendor.
Where bargaining power sits, in one sentence: chip vendors fix the spec, hyperscalers control the BOM, ODMs hand the spec down, and Auras lives or dies on whether it qualifies before the alternative does. The one place where it has real pricing power is on new liquid-cooling content with low-volume specialised tooling — cold plates, CDU/RPU integration, quick disconnectors brought in-house since 2012 via an IBM tech transfer. Everywhere else, it is a price-taker.
2. The Playing Field
Auras is the smallest of the dedicated Taiwan thermal-module pure-plays and sits one tier below the integrated power/thermal giants and the US data-centre infrastructure leaders. The peer table that matters is short.
Vertiv revenue and market cap shown above are US-dollar figures translated to NT$ at the current rate for like-for-like comparison; Vertiv reports in USD. Jentech (3653.TW) was added late in this run from the Yahoo chart endpoint — only market cap is reliable; income data is not yet staged.
The map tells the real story. Auras is the lowest-operating-margin name in the set and the smallest by market cap, despite booking the second-highest revenue growth (+47% FY24→FY25 vs AVC +95%). Two readings of that.
The bullish read: there is operating leverage still locked up in mix shift. AVC, the closest peer, runs operating margin almost 600bp higher on a similar-margin gross profit line — meaning Auras has room to widen if liquid-cooling volume catches its tooling ramp. The bearish read: Auras has the worst scale economics in the group and lacks AVC's customer relationship depth, which is what shows up in the AVC ROE of 49.9% versus Auras 23.2%. The valuation reflects this — Auras trades at the lowest P/E (~38x) and lowest P/B among the listed thermal-cluster names with full financials, not at the kind of multiple AVC and Vertiv command. The market is paying for the cycle, not for the franchise.
What "good" looks like in this industry is not gross margin — it is gross margin × revenue per FTE × capex turn. AVC's blueprint (NT$15B 2026 capex against NT$140B revenue, ~9% capex intensity) is the productivity benchmark to watch Auras against; Auras spent NT$2.30B against NT$23.3B in FY2025 (~10% capex intensity) but has not yet shown the AVC-style revenue throughput per NT$ of installed capacity.
3. Is This Business Cyclical?
Yes — and in two different ways at the same time. There is a chip-platform cycle (AI server TDPs and hyperscaler capex) and a commodity-component cycle (PC/notebook unit shipments and copper/aluminum prices). FY2018 is the right historical anchor because it shows what the business looks like when both cycles are in trough mode at once.
Where the cycle hits first. Working capital — accounts receivable jumped from NT$5.4B (FY24) to NT$10.1B (FY25), inventory from NT$2.8B to NT$6.1B. Operating cash flow swung from +NT$1.64B to ‑NT$0.57B, even though net income rose 33%. Capex doubled to NT$2.30B. The combined result was the FCF margin going from +1.0% to ‑12.4%. The capital cycle for this business turns before the P&L, not with it: when AI server orders accelerate, the company books revenue and net income but absorbs cash into receivables, inventory and tooling. The mirror image happens on the way down — cash converts faster than reported profit reverses, which is what makes FY2018 a brutal but survivable trough year.
The leading signal investors should watch is Auras's own monthly revenue disclosure on the 10th of each month (TPEx requirement). Q1 FY2026 printed NT$8.55B vs Q1 FY2025 NT$4.41B = +94% YoY. April 2026 was NT$2.84B vs April 2025 NT$2.02B = +41%. Sustained 40-50% YoY through the rest of FY26 would lift the revenue base to ~NT$35B — but a deceleration below 20% within two months would be the first hard signal that the AI-platform cycle is rolling over.
4. The Metrics That Actually Matter
The non-obvious read is the gap between gross margin and FCF margin: a 27.4% gross margin printed alongside a ‑12.4% FCF margin means the entire reported P&L profit, and more, is being recycled into next-platform capex and into receivables/inventory build. That is the right answer in a ramp. It would be a red flag if AI server demand failed to show up to consume the capacity — and the monthly revenue line is the place to verify it is showing up. So far, it is.
Three metrics are not on this list and should be deliberately ignored as the headline story: trailing P/E (the cycle distorts the denominator), dividend yield (~0.9% — irrelevant to the thesis), and book value (this is an income-statement story, not a balance-sheet one).
5. What Is This Business Worth?
Auras is best valued as one normalised earnings stream tied to AI-server thermal content, discounted by the risk that the franchise has to be re-won every NVIDIA platform generation. It is not a sum-of-the-parts story — there is one reported segment and one customer set. There is no holding-company structure, no listed subsidiary, no separable real-estate or financial-services drag.
The right way to think about value is: (steady-state revenue at the next platform mix) × (steady-state gross margin) × (steady-state opex ratio) × (re-bid probability per platform). The market today is implicitly underwriting all four favourably.
The lens the multiple should be built around. At NT$1,065 share price and 91.0M basic shares, market cap is ~NT$97B, on FY25 net income of NT$2.57B — a trailing P/E of ~38×. FY25 EPS of NT$28.26 implies the market expects a roughly 50% earnings uplift in the next 12-18 months to reach a forward P/E in the low-20s on out-year consensus. That is consistent with the Q1 FY26 monthly revenue run-rate (+94% YoY) holding for half the year and gross margin staying near 27%. The right valuation lens is therefore a 1-year forward earnings number anchored to the monthly revenue trajectory, sanity-checked against AVC's multiple at 53.7×. Auras at 38× and AVC at 53× says the market is paying a discount for Auras's slower scale ramp and weaker ROE — that discount is fair if the next platform re-bid is meaningfully more contested.
What would make the multiple look cheap in hindsight: monthly revenue holding >50% YoY through 2H 2026, gross margin staying >27%, FCF turning positive in FY27, top-3 customer share staying in the 30-40% band. What would make it look expensive: a single monthly revenue print below +20% YoY, gross margin compressing below 25%, FCF margin still negative two years from now, or top-3 customer share moving above 50%.
6. What I'd Tell a Young Analyst
Read the monthly revenue print on the 10th of every month before you read anything else about this company. It is the most leading indicator of cycle state in Taiwan tech and it cannot be window-dressed. Anything below +20% YoY for two consecutive months is the first piece of evidence that the FY2018 base case is back in play.
Do not mistake the gross-margin expansion since FY2018 for structural progress. It is mix-driven. Liquid-cooling content is the cycle override; the underlying air-cooling business is a price-taker. Track the FY2026 annual report disclosure of the AI server / NB / VGA / Auto split — the 39 / 39 / 17 / 2 split published for FY2024 is the cleanest version of the mix question.
Do not flinch at the negative FCF — flinch at the wrong reason for it. NT$2.30B of FY25 capex is acceptable if it earns back at >25% gross margin on next-platform liquid-cooling volume. It is unacceptable if it earns back on legacy air-cooling at low-teens gross margin. The forensic flag is the OCF/NI ratio of ‑0.22x — that is working-capital absorption on top of capex, and you want to see it pivot above 0.5x within four quarters as the AI server ramp converts to cash.
The market may be overestimating gross margin durability past FY2027 (AVC's NT$15B+ capacity ramp is positioned to recalibrate cold-plate pricing) and underestimating the durability of the demand pull from NVIDIA's >1,000W AI accelerator roadmap. The interaction between those two errors defines the next 18 months.
What would change my mind on the bull case: monthly revenue YoY printing below +20% for two consecutive months without a one-off explanation, AVC capacity ramp slipping by 2+ quarters (would help Auras pricing), the top-3 customer share moving above 50% (single-platform dependency), or a gross-margin print below 24% (legacy pricing wins).
What would change my mind on the bear case: an automotive design-win at Level 3+ ADAS, FCF margin turning positive in FY26 (faster than current trajectory implies), top-3 customer share falling below 30%, or a confirmed 2-phase immersion-cooling design win at a Tier-1 hyperscaler.