History
Figures converted from TWD at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.
The Auras Story — From Notebook Cooler to AI Server Liquid Cooling
Across seven annual reports, Auras evolved from a 12% gross-margin notebook-cooler subcontractor (FY2018) into a 27% gross-margin AI server liquid-cooling vendor (FY2025). The transformation came in three discrete waves — a margin reset in FY2019 (5G + server), a product-portfolio pivot in FY2022–FY2023 (liquid cooling, CDU/manifold/quick-disconnector), and a commercial-scale AI server ramp in FY2024–FY2025. Management's directional calls — on AI server growth, liquid cooling becoming mainstream, and 2025 profitability — have largely been validated by the numbers; but the risk-factor boilerplate has lagged badly behind the company's actual capex commitments, customer concentration, and balance-sheet activity. Credibility is high on strategy; lower on disclosure hygiene.
1. The Narrative Arc
The story didn't change once — it changed three times. FY2019 transformed the margin. FY2022–FY2023 redefined the product. FY2024 monetised it. Each pivot showed up in the language of the annual report 12–24 months before it showed up in the income statement.
2. What Management Emphasized — and Then Stopped Emphasizing
Annual-report emphasis (0 = absent, 5 = central thesis).
Three topics rose, three faded, two stayed flat:
- 5G smartphone / vapor chamber went from absent (FY2018) to the centerpiece of the consumer-facing pitch (FY2020–FY2021) to literally deleted from FY2024's product-trend section — the report writes "(Smart phones deleted)" as a parenthetical. The vapor-chamber R&D didn't disappear; the smartphone customer story did.
- Bitcoin mining and metaverse show up exactly once each (FY2021) as growth tailwinds and are gone by FY2023. These were demand cushions, not strategy.
- Liquid cooling is the inverse — first appears as one of nine R&D items in FY2021, becomes a "solution" by FY2022 (with the "IBM tech transfer 10 years ago" framing), and is the dominant productised offering by FY2024 (cold plates, pumps, CDU, manifolds, quick disconnectors all named).
- AI server as a discrete category is brand new in FY2022 — its first appearance cites ChatGPT and a TrendForce projection that AI servers will be 15–20% of server value in 2023 rising to 20–30% by 2024–2025.
- Auto / EV thermal has been promised every year since FY2018 and remains 2% of revenue in FY2024 — the canonical dropped initiative that nobody flags as dropped because management never raises it loudly.
The metaverse and crypto-mining mentions in FY2021 are the clearest example of management chasing a narrative they already knew was fragile. Both were deleted within two reports, with no acknowledgment.
3. Risk Evolution
Disclosed risk emphasis (0 = absent, 5 = prominent).
The risk-factors section is structurally a boilerplate template — the same 13 categories appear every year with near-identical wording. The substantive changes are visible only by reading between the lines:
- Customer concentration flipped from "no customer above 10%" through FY2022 to three customers above 10% by FY2024 (Q ~13.6%, D ~11.6%, T ~10.7%). Customer T jumping to 10.7% is explicitly attributed to "increase in sales of liquid and air cooling for new models of servers" — the AI server ramp creating the very concentration risk the prior boilerplate denied.
- "No plans for expanding plants currently" appears verbatim in FY2018, FY2019, FY2022, FY2023, and FY2024 — even though FY2024's MDA simultaneously discloses construction of a new corporate headquarters, new factories, and a convertible-bond issuance to fund it. The risk-factor template stopped describing reality somewhere around FY2020 (Thailand plant) and never caught up.
- Currency was the loudest disclosed risk early on (FY2020 carried a $6M exchange loss). It quietly improved as TWD weakness turned exchange-loss line items into gains (FY2022 +$9.5M; FY2024 +$9.3M). Same business; reversed sign.
- China manufacturing exposure is the most under-disclosed risk in the entire file. Roughly 80–90% of purchases are related-party (Mainland China subsidiaries) and Hefei/Southwest sites still dominate output. The Thailand and (FY2024) US/Vietnam diversification narratives are in the business section, not the risk-factor section.
The boilerplate "no plans for expanding plants currently" line in the FY2024 risk-factor section, written in the same year management raised convertible-bond financing for new factories and a new corporate HQ, is the single clearest disclosure-hygiene issue across this seven-year file.
4. How They Handled Bad News
There are two episodes of bad news worth examining: FY2022 (revenue down 2.8%) and FY2023 (revenue down 8.3%, two consecutive years of decline).
The FY2022 response was almost honest — management acknowledged the revenue dip but pointed (correctly) to record profit, without flagging that the bottom-line record was partly an FX accident. The FY2023 response was the more telling one: rather than apologise for two straight years of revenue decline, the report reframed Auras as a liquid-cooling solution provider that had just shed lower-margin business. The GM expansion (19.6% → 23.6%) made the reframing land. This is the only time in the file where management successfully turned a top-line miss into a positive story without papering over it.
The FY2024 episode is different: there is no bad news in the income statement (revenue +24%, NI +56%) — but management buried the convertible-bond issuance, the new-headquarters capex, and the customer-concentration spike across separate sections of the report. The risk-factor section was not updated to reflect any of it.
5. Guidance Track Record
Auras does not provide formal quantitative guidance the way US-listed companies do. The promises worth tracking are the directional claims management embeds in their forward-looking discussion — and a few specific external citations they leaned on.
Credibility Score (1–10)
Why 7.5/10. The directional calls — AI server growth, liquid cooling becoming the mainstream DC cooling solution, 2025 being a profitability inflection — were not just correct but understated. AI server value share is higher than management's 2022 projection. Liquid cooling is already mainstream. The 2025 profit step-up promised in the FY2024 report delivered with revenue +47% and EPS +33%. Where management lost a point is the slightly-too-eager consumer-narrative claims (5G phones outrunning 4G by 53.7%, PUE going below 1.1, AI PC replacement cycle in 2024) — defensible but premature. The remaining 2.5-point gap is disclosure: the risk-factor section is structurally divorced from the actual capital-deployment and concentration realities visible elsewhere in the same filing.
6. What the Story Is Now
Auras today is a thermal-management vendor whose product mix, customer mix, geographic mix, balance sheet, and headcount have all changed in the same direction over the last 24 months: toward AI server liquid cooling for US/EU hyperscaler-adjacent customers. The numbers validate the pivot — revenue went from $445M (FY2023) to $815M (FY2025) and gross margin from 23.6% to 27.4%. R&D headcount doubled from 240 to 486. Americas exposure went from ~1% to 14%. The narrative was led by the income statement, not the other way around.
FY2025 Revenue ($M)
FY2025 Net Income ($M)
FY2025 Gross Margin
FY2025 EPS ($)
What has been de-risked. The product-portfolio question (can Auras actually build cold plates, CDUs, manifolds, and quick disconnectors at scale?) is closed — three customers are now buying enough of it to each exceed 10% of revenue, and the Americas/Europe mix proves the customers are real. The margin question (was the FY2023 GM expansion a one-off?) is also closed; FY2025 GM 27.4% is the highest in the file.
What still looks stretched. Three things:
- Customer concentration: 36% of FY2024 revenue from three customers, almost certainly higher in FY2025 given the monthly revenue acceleration. Auras's customers are the same hyperscalers everyone else's customers are.
- Capex pace vs disclosure: FY2024 saw convertible-bond issuance, a new corporate HQ, new factories, and equity acquisitions of Zhongyangguang and Taiwan Metal Precision — but the risk-factor section still reads "no plans for expanding plants currently." When the boilerplate stops matching reality, the next material disclosure surprise tends to come from the same gap.
- Margin durability: the 4pp GM expansion FY2022→FY2023 happened during a revenue decline — meaning Auras shed lower-margin business and the remaining higher-mix products lifted the average. As the AI server volumes scale into FY2026, the more relevant question is whether Auras can hold 25%+ GM with a far more competitive vendor landscape (Asia Vital Components, CCI, Forcecon, plus Foxconn) targeting the same hyperscaler sockets.
What to believe vs discount. Believe the strategic positioning — the R&D headcount, customer wins, and geographic mix shifts are all real and visible in the source documents. Believe the FY2025 profitability print. Discount the boilerplate risk-factor section (it has not described reality since 2020) and discount any new tangential narrative (auto/EV thermal has been promised every year since 2018 with no observable revenue translation). The current story is simpler than the 2021 version (no metaverse, no crypto, no 5G phone) but more concentrated than it was — and that concentration is the durable risk for the next two years.